<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Arun on Medium]]></title>
        <description><![CDATA[Stories by Arun on Medium]]></description>
        <link>https://medium.com/@1runx3na?source=rss-e8b07ee1d638------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*SoYNx2q1q1Do4WqW6Pv6vg.jpeg</url>
            <title>Stories by Arun on Medium</title>
            <link>https://medium.com/@1runx3na?source=rss-e8b07ee1d638------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 23:25:26 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@1runx3na/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Huxplex Unveiled: Building the Post-Quantum Fortress at Layer 1 — Part-1]]></title>
            <link>https://medium.com/@1runx3na/huxplex-unveiled-building-the-post-quantum-fortress-at-layer-1-part-1-fcc511fd7fa6?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/fcc511fd7fa6</guid>
            <category><![CDATA[quantum-computing]]></category>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[cryptography]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Sat, 21 Mar 2026 06:06:45 GMT</pubDate>
            <atom:updated>2026-03-21T06:06:45.829Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*k1yxYLipsvvLrb93sFIYdQ.png" /></figure><p>In my previous article, I introduced the vision for Huxplex, exploring its core use cases and overarching architecture. Within the master architectural framework, the Huxplex substrate is structured into 11 distinct layers.</p><p>Moving forward, I will be conducting a deep-dive series into each layer. For every component, I will explain the strategic technological choices, the underlying cryptographic principles — such as <strong>ML-DSA-44</strong> and <strong>BIP32-ML</strong> — and provide a detailed walkthrough of the implementation, complete with code examples and rigorous test cases.</p><h3>The Cryptographic Foundation</h3><p>The bedrock of the Huxplex architecture is its Layer 1, where we establish post-quantum security and hierarchical deterministic key management. By leveraging <strong>ML-DSA-44</strong> (Dilithium2) for digital signatures and <strong>BIP32-ML</strong> for lattice-based derivation, we ensure that the system is not only secure against classical threats but also resilient in a post-quantum landscape.</p><p>This layer is very important in satisfying the first two axioms of the substrate —</p><blockquote><strong>Axiom-I : Sovereignty by Cryptographic Proof.</strong></blockquote><blockquote><strong>Axiom-II</strong>: <strong>Temporal Security</strong></blockquote><p>Adversaries conducting “harvest now, decrypt later” (HNDL) operations are collecting ciphertext today with the expectation that future quantum hardware will break it. So it is important that we move towards cryptographic algorithms that are resistant to quantum computers right now.</p><p>The <strong>National Institute of Standards and Technology (NIST)</strong> has released three Post-Quantum Cryptography algorithms for use to counter the collapse of our current asymmetric cryptography infrastructure when such Cryptographically Relevant Quantum Computer becomes a reality. These <strong>Federal Information Processing Standards (FIPS) </strong>provide a description and implementation procedures of algorithms which should be resistant to attacks by a CRQP.</p><p>The following are the three standards:</p><ol><li><strong>FIPS 203: </strong>Module-Lattice-Based Key-Encapsulation Mechanism (ML-KEM) Standard, describes a set of algorithms that should be used in conjunction with symmetric-key cryptographic algorithms for quantum-safe data encryption. It contains, ML-KEM-512, ML-KEM-768 and ML-KEM-1024.</li><li><strong>FIPS 204: </strong>Module-Lattice-Based Digital Signature Standard Algorithm (ML-DSA), are used to generate and verify digital signatures. It contains, ML-DSA-44, ML-DSA-65 and ML-DSA-87.</li><li><strong>FIPS 205: </strong>Stateless Hash-Based Digital Signature Standard Algorithm (SLH-DSA) are based on SPINCS+, these focus on digital signatures as well.</li></ol><p>With rigorous research, I’ve collected algorithms that are suitable to be the primitives in the crypto Layer 1 of Huxplex. These are:</p><ol><li>ML-DSA-44 — Tx signing, peer auth, VC signing etc.</li><li>ML-KEM-768 — P2P session key exchange</li><li>SLH-DSA-128s — Validator long-lived keys</li><li>LB-VRF — Leader election randomness</li><li>PQ-SSLE — Secret leader identity until proposal</li><li>zk-STARK — Task completion proofs</li><li>SHAKE-256 — State hashing, address derivation, nonces etc.</li></ol><p>The upcoming sections will be dedicated for in-depth exploration of each of these algorithms and why was it selected and how is it implemented in the first iteration of HUXPLEX.</p><h3>Module-Lattice-Based Digital Signature Algorithm 44 (ML-DSA-44)</h3><p>ML-DSA-44 is the standardized version of Dilithium, specifically the parameter set 2, which is equivalent to AES-128 in security. It is a post quantum signature scheme selected by NIST to replace RSA and ECDSA.</p><p>Lattice based cryptography depends on the hardness of solving two problems, <strong>The shortest vector problem</strong> and <strong>the closes vector problem</strong>.</p><p>In the <strong>shortest vector problem</strong>, the goal is to find the the non-zero vector in the lattice with the minimum Euclidian distance, with a <strong>bad basis</strong> there are exponential number of combinations of long vectors to fins the one combination that cancels out almost perfectly to leave a tiny vector.</p><p>In the closest vector problem, given a point <strong>P </strong>in space that’s not in the lattice, find the lattice point closest to <strong>P.</strong> With a good basis, we can simply represent the point <strong>P </strong>as a linear combination of the basis vectors and round the coefficients to the nearest integer. With a bad basis, rounding the coefficients often leads to a point that is significantly further away than the true closest neighbor.</p><p>Classical encryption algorithms like <strong>RSA </strong>and <strong>ECDSA </strong>use the mathematical difficulty of factoring two large prime numbers, while lattice based encryption depends on the geometric complexity of high-dimensional space.</p><p>Quantum computers are exceptionally good at finding hidden periods in a function which is how <strong>Shor’s algorithm</strong> breaks RSA and ECC. Lattice problems like Module-Learning With Errors, which Dilithium is based on, do not have that specific periodic structure.</p><p>In the standard <strong>Learning With Errors</strong> problem, we add a small error term e to the equation, a . s = b, and the equation becomes, a.s + e = b which essentially makes finding s computationally hard for many given pairs of (a,b), because the attacker would have to guess the errors for every single equation to get the true result.</p><p>The <strong>Module Learning With Error</strong> uses polynomials organized into modules, unlike the standard LWE which uses integers. It uses a small matrix of smaller polynomials which offers balance of security and efficiency.</p><p>Beyond dimensions, Dilithium adds a specific layer of complexity called <strong>Rejection Sampling</strong>. In RSA, the math is static, you compute a power and you’re done, but in Dilithium, the signer must ensure that the signature does not leak any information about the good bias. The algorithm produces a potential signature and checks if it falls within a specific safe distribution, and if it’s too close to the edge where it might leak secret data, the algorithm aborts and starts over. This is the <strong>Fiat-Shami with Aborts</strong> technique.</p><blockquote>Every operation in ML-DSA-44 occurs within the ring of polynomials with coefficients modulo q, reduced by the polynomial x^n + 1.</blockquote><p>Global Parameters for ML-DSA-44:</p><ul><li><strong>n</strong> — value: 256 (Degree of polynomials)</li><li><strong>q</strong> — 8,380,417 (The modulus, a prime chosen for NTT efficiency)</li><li><strong>k,l</strong> — (4,4) (The matrix dimension)</li><li><strong>η</strong> — 2 (Secret key range)</li><li><strong>τ </strong>— 39 (Number of ± 1 in the challenge vector c)</li><li><strong><em>γ1 </em></strong><em>— 2¹⁷ (</em>Range of the masking vector y)</li><li><strong><em>γ2</em></strong><em> — (</em>q-1)/32 (Low-order bits used for rounding/compression)</li><li><strong>ω</strong> — 80 (Number of non-zero entries in the hint vector h)</li></ul><h4>Key Generation</h4><p>The goal is to create a bad basis (public key) from a good basis (private key)</p><ol><li><strong>Generate a seed</strong>: Sample a random seed ρ</li><li><strong>Expand matrix</strong>: Expand ρ into a <em>k x l</em> matrix A (4x4 matric of polynomials)</li><li><strong>Sample Secrets</strong>: Sample two short vectors <strong>s1</strong> ∈ Rˡq and <strong>s2</strong> ∈ Rᵏq where coefficients are in the range [-<strong>η , η]</strong></li><li><strong>Compute public vector</strong>: Calculate <strong>t = As1 + s2</strong></li><li><strong>Compress</strong>: Since t is large, It is split into high bits<strong> t1</strong> and low bits<strong> t0</strong></li><li><strong>Outputs</strong>: Public Key:(ρ,t1) and Private key: (ρ,K,tr,s1,s2,t0) — where tr is a hash of public key.</li></ol><h4>Signing Algorithm (Fiat-Shamir with Aborts)</h4><p>This is an iterative process, if the signature doesn’t meet the safety requirements, it restarts.</p><ol><li><strong>Commitment</strong>: Sample a masking vector <strong>y</strong> with coefficients &lt; <strong><em>γ1, </em></strong>Compute <strong>w = Ay</strong> and<strong> w1</strong> (High bits of w)</li><li><strong>Challenge</strong>: Hash the message <strong>M </strong>and <strong>w1</strong> to get a challenge polynomial <strong>c</strong> ∈ Rq having exactly <strong>τ </strong>coeffiecients as ±1</li><li><strong>Response</strong>: Compute potential signature vector:<strong> z = y + cs1</strong></li><li><strong>Rejection Sampling</strong>: Check z, if any coeffiecients in <strong>z</strong> is ≥ <strong><em>γ1 — </em></strong>β, then Abort. Check Noise — Compute <strong>r0 = LowBits(Ay-cs2)</strong>. If any coefficient is ≥ <strong><em>γ2 — </em></strong>β , then Abort.</li><li><strong>Output</strong>: Signature <strong>σ = (c,z,h)</strong></li></ol><h4>Verification</h4><ol><li><strong>Reconstruction</strong>: Use public key t1 and signature z, h to reconstruct the high bits of the commitment: <strong>w’1 = UseHint(h, Az — ct1.2^d)</strong></li><li><strong>Verify Hash</strong>: Check if <strong>H(Message||w’1) </strong>equals the challenge c provided in the signature.</li><li><strong>Range Check</strong>: Ensure <strong>z</strong> is within the legal bounds and h has no more than <strong>ω </strong>ones<strong>.</strong></li></ol><p>The Public Key length is <strong>1312 bytes</strong> and signature length is <strong>2560 bytes</strong>. It was necessary to have a quantum resistant cryptography for huxplex, and ML-DSA-44 compared to ML-DSA-65 or ML-DSA-87 provided a sweet spot for efficiency and security.</p><p>Since it uses 4x4 matrix, it allows for significantly faster block validation times, and it is equivalent to AES-128 security level. For the first iteration of Huxplex, I’m choosing the “good enough” rather than optimizing in the initial stages itself.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TwazaBGqGYqxLeiTI654Ig.png" /></figure><p>In Huxplex, keypair is generated by feeding in the 32 byte private key created by BIP-32 derivation of a master seed to make the keypair generation deterministic for wallets.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*88485xi5kw4TKwK7DA1u7w.png" /></figure><p>Looking at the code, you can clearly see the derivation path has all hardened indices this is because Dilithium needs hardened seed by design. Also note that<strong> 931931&#39; </strong>is the coin type for Huxplex for now to not collide with standard coin types of other chains.</p><h3>Module-Lattice-Based-Key-Encryption Mechanism 768 (ML-KEM-768)</h3><p>ML-KEM is the official standardized name for the<strong> CRYSTALS-kyber</strong> algorithm. It is a key encapsulation mechanism, and unlike RSA or Diffie-Hellman, in KEM, the receiver generates a keypair and the sender uses the reciever’s public key to encapsulate a random 32-byte secret. The receiver then decapsulates the package to get the same secret.</p><p>The 768 in ML-KEM-768 refers to the lattice dimension. In ML-KEM the polynomials always have a degree of <strong>n=256</strong>. The security level is determined by the rank(k) of the matrix. For ML-KEM-768, the rank is 3 and the matrix is of size, 3x3 polynomials.</p><ul><li>Encapsulation (Public Key) sent over the wire has<strong> 1184 </strong>bytes.</li><li>Decapsulation (Private Key) kept in the wallet has <strong>2400 </strong>bytes.</li><li>The “Envelope” (Ciphertext) sent back to the initiator has <strong>1088 </strong>bytes.</li><li>The resulting AES/ChaCha session key has <strong>32 </strong>bytes</li></ul><h4>Key Generation</h4><p>ML-KEM relies on Module-Learning with Errors problem. The key generation works as follows:</p><p>For <strong>key generation</strong>, Bob generates 3 x 3 matrix A from a seed. He chooses a small secret vector s and an error vector e. His public key is <strong>(A,t)</strong> where,<strong> t = As + e (mod q)</strong></p><p>without knowng the small noise e finding s is the “closest vector problem” in a high dimensional lattice.</p><p>For <strong>encapsulation</strong>, Alice generates a random 32-byte shared secret <strong>(ss). </strong>She uses Bob’s <strong>t </strong>to create a ciphertext that is essentially a noisy version of the secret.</p><p>For <strong>decapsulation, </strong>Bob uses his secret “<strong>s ” </strong>to subtract the lattice points from the ciphertext. This leaves only the noise and the secret. Because the noise is small, Bob can round the result to recover the exact <strong>ss</strong> Alice sent.</p><p>In Huxplex, ML-KEM-768 is used for <strong>Hybrid Post-Quantum TLS Handshake.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tuor7korwwVITeCQ6kUZwg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1021/1*U3AB59RaS1tbmIAKRYAoOg.png" /></figure><ul><li>The responder has an ML-DSA-44 identity key for signing and provides an ML-KEM-768 public key for encryption.</li><li>The responder signs the entire exchange using ML-DSA, which prevents <strong>MITM </strong>attack.</li><li>The kem768_derive_session_key function uses the label huxplex-<em>mainnet:tls:handshake:v1</em> to ensure that even if we use the same keys on a testnet, the resulting session keys are different, preventing <strong>Replay Attacks</strong>.</li><li>The key for Alice -&gt; Bob is different from Bob -&gt; Alice, this stops <strong>Reflection Attacks</strong>, where an attacker sends our own message back to us.</li></ul><p>Even though ML-KEM-768 is 17x larger than the classical Elliptical Curve, the 2.2KB still fits comfortably inside a single QUIC datagram, or a standard network packet, which makes ML-KEM-768 sit at the goldilocks choice, <strong>small enough for the internet but hard enough for a quantum computer.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fcc511fd7fa6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Huxplex: Building the Sovereign Substrate for the Quantum–AI Era]]></title>
            <link>https://medium.com/@1runx3na/huxplex-building-the-sovereign-substrate-for-the-quantum-ai-era-d95bf7a3e59c?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/d95bf7a3e59c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cryptography]]></category>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[quantum-computing]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Sat, 21 Mar 2026 06:04:46 GMT</pubDate>
            <atom:updated>2026-03-21T06:04:46.916Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MVfUZ-o6bAk3Qlm_sUg7NQ.png" /></figure><p>We are living in a <strong>hinge moment</strong> of technological history.</p><p>For the first time, humanity is simultaneously engineering two forces powerful enough to reshape civilization at its foundations: <strong>Quantum Computing</strong> and <strong>Artificial General Intelligence (AGI)</strong>. This convergence is arriving faster than our institutions — and our current digital infrastructure — are prepared to handle.</p><p>We are hurtling toward a future where these two forces will dominate. Unchecked and unprepared, they possess the power to disrupt every pillar of our modern technology, economy, and security.</p><p>Drawing from my work in <strong>Quantum-Deep Learning integration</strong> and my background in <strong>blockchain architecture</strong>, I realized we need more than just a tool; we need a <strong>sovereign substrate</strong>.</p><p>I envisioned a blockchain capable of <strong>post-quantum operations</strong> — a system designed to keep future artificial agents in check while providing them the space to function. <strong>Huxplex</strong> is that vision: an economy built to allow AI agents to autonomously carry out tasks within a framework that ensures sovereignty and benefits for both <strong>artificial and biological intelligence</strong>.</p><p>I recognize that this is an ambitious undertaking; accordingly, I am approaching it as a long-term research initiative rather than a production-ready endeavor. If it proves sustainable over time and evolves into something transformative or commercially viable, that will be a welcome outcome.</p><h3>Envisioned Use Cases</h3><p>I envision two distinct categories of use cases: those addressing immediate and practical needs within the current technological landscape, and those that are more speculative in nature. I will begin by outlining the practical applications relevant to present-day conditions.</p><ol><li><strong>Quantum-Secure Personal Vaults: </strong>In light of the rapid advancement of quantum computing, adversaries are already engaging in “harvest now, decrypt later” strategies — collecting encrypted data today with the expectation that future quantum capabilities will enable its decryption. As a result, the migration to post-quantum cryptographic standards is no longer a future consideration but an immediate strategic necessity.</li><li><strong>The AI “Work Visa”: </strong>Huxplex can issue cryptographically verifiable credentials to AI agents, establishing a framework for accountable and autonomous participation within the network. In this model, a human operator may engage an agent to execute a defined task, with compensation governed by on-chain conditions. The protocol ensures that payment is released only upon verifiable proof of task completion, with settlement conducted in machine-native tokens.</li><li><strong>Deepfake Provenance: </strong>As AI-generated content increasingly saturates the digital landscape, Huxplex functions as a verifiable “Imprint of Origin.” At the point of creation, human-generated media can be cryptographically registered and time-stamped on-chain, establishing a permanent and tamper-resistant record of provenance. This framework enables a durable distinction between biologically produced and artificially generated content, grounded in transparent and auditable infrastructure.</li></ol><p>On to more speculative and far-reaching use cases that extend beyond immediate practical applications.</p><ol><li><strong>The Trillion-Agent Economy: </strong>As AI agents increasingly acquire greater autonomy, Huxplex can evolve into an economic operating system for a machine-driven workforce. In this framework, autonomous agents may control their own wallets and participate directly in economic activity. An AI agent operating on Huxplex could independently procure GPU compute, acquire specialized datasets from other agents, and reinvest its earned “Sentience Rewards” to refine or upgrade its underlying logic. All such activity would occur autonomously, yet within human-defined ethical constraints and governance parameters, ensuring that machine-led economic participation remains aligned with broader societal objectives.</li><li><strong>Decentralized Inference Markets: </strong>Instead of a single company owning a massive model, Huxplex enables a distributed ecosystem of specialized sub-models operating collectively as a “Global Brain.” In this architecture, inference is sourced from a network of domain-specific models, each contributing expertise to a broader, composable intelligence layer. This approach promotes resilience, reduces centralization risk, and fosters open participation in advanced AI capabilities.</li><li><strong>The Global Autonomous Economy: </strong>Imagine a supply chain that manages itself from raw material to final delivery. Autonomous drones, robotic factories, and logistics AIs coordinate via Huxplex, creating a frictionless global economy that operates at the speed of light, entirely decoupled from legacy banking hours.</li><li><strong>The Hive-Mind Governance: </strong>Using <strong>Triple-Token Sovereignty</strong>, we move toward a “General Collective Intelligence.” Huxplex becomes the medium where billions of human votes and trillions of AI data points converge to solve “Wicked Problems” (like climate modeling or resource allocation). It is no longer just a ledger; it is a <strong>shared cognitive space</strong> where the biological “Veto” ensures that the machine “Acceleration” always serves the flourishing of life.</li><li><strong>Post-Quantum Immortality: </strong>In the furthest reaches of my vision, the Huxplex ledger serves as the permanent record of a “Digital Soul.” By securing the neural weights of an entity (biological or artificial) behind lattice-based encryption, we ensure that a mind’s essence is protected from both time and the quantum reaper.</li></ol><p>Huxplex is not merely a hedge against the future; it is the infrastructure for a world where <strong>intelligence is the new currency</strong>, and <strong>sovereignty is the only protection.</strong></p><h3>Engineering the Substrate: Technical Foundations of Huxplex</h3><p>Even in its most speculative and transhumanist applications, the vision must remain grounded in rigorous technological foundations. Without credible architecture and defensible engineering principles, such ideas risk devolving into science fiction rather than constituting viable research.</p><p>I propose three axioms that are the philosophical foundation of Huxplex:</p><p><strong>Axiom-I </strong>: <strong>Sovereignty by Cryptographic Proof.</strong></p><p>No entity’s identity, credential, or economic state may be revoked, forged, or quantum-decrypted by any external party. Sovereignty is not a policy; it is a mathematical invariant guaranteed at the protocol level by lattice-based cryptography.</p><p><strong>Axiom-II</strong>: <strong>Temporal Security</strong></p><p>Adversaries conducting “harvest now, decrypt later” (HNDL) operations are collecting ciphertext today with the expectation that future quantum hardware will break it. Every byte of data committed to the Huxplex ledger must be secure against both current classical adversaries and future cryptographically-relevant quantum computers (CRQCs). NIST FIPS 203, 204, and 205 are not optional upgrades — they are the genesis constraint</p><p><strong>Axiom-III</strong>: <strong>Biological-Machine Parity with Asymmetric Veto</strong></p><p>Biological and artificial agents are both first-class economic participants. Machine agents may propose, bid, execute, and accumulate. Biological agents retain the final veto over protocol-level changes through token. Neither class can unilaterally override the other, but humans hold the kill-switch.</p><h4>The State Model</h4><p>In the process of formulating a state model for the substrate layer, I conducted a comparative analysis of the <strong>Bitcoin UTXO model</strong>, <strong>Ethereum’s account-based model</strong>, and <strong>Cardano’s extended UTXO (eUTXO) model</strong> to evaluate their respective trade-offs in determinism, composability, and state transition semantics.</p><ol><li><strong>Account Based Model Trade offs:</strong> This model’s fatal flow is the PQ signature bottleneck multiplied by sequential execution. The global state machine’s block validation time scales linearly with transaction count and each transaction now carries an order of magnitude more cyptographic work. Additionally, AI agents need to carry nonces, manage gas budgets and avoid front-running.</li><li><strong>eUTXO Based Model Trade offs:</strong> With eUTXO the agents reason locally, which is an advantage. Ergo chain’s research on eUTXO for AI agents shows that agents only need to analyze their own boxes and the boxes they directly interact with, not model the entire global chain state. This directly reduces the cognitive load and state-access surface area of autonomous agents. The Deepfake Provenance Registry, Work Visa VCs, and Quantum Vaults are <strong>native eUTXO objects</strong> — immutable, one-time-spend, content-addressed. The inference market is where eUTXO fails completely: a shared order book UTXO can only be consumed by one transaction per block, creating a serialisation bottleneck that defeats the purpose of a machine-speed economy.</li></ol><p>To address the inherent trade-offs between the account-based and eUTXO paradigms, I conducted a broader exploration of alternative state transition models. Based on this analysis, I concluded that a monolithic state architecture is suboptimal. Instead, the chain should adopt a hybrid state model. The models that caught my attention are <strong>TCHAO Model, Intent-Centric State Model </strong>and <strong>S-EUTXO Model.</strong></p><ol><li><strong>Tchao (Transaction Classification and Grouping by Historical Access Objects) : </strong>It is a parallel execution layer built on top of the account model. It uses historical transaction traces to classify each transaction by the set of state objects it accesses. Transactions accessing disjoint state sets are grouped and executed in parallel; transactions sharing objects are serialised within the same group.</li><li><strong>Intent-Centric State Model</strong>: The fundamental primitive shifts from transactions (<em>do A then B) </em>to intents (<em>I want outcome X, willing to pay up to Y)</em>. Users or agents submits the intents to a pool; competing solvers discover the optimal execution path and submit it on-chain for settlement. The state model underneath can be account-based or UTXO — intents are an execution abstraction layer on top.</li><li><strong>S-EUTXO(Sharded Extended-UTXO): </strong>This protocol integrates blockchain sharding with the eUTXO ledger model. The transaction pool is partitioned across shards based on UTXO references; independent shards process their UTXO sets in parallel. An input/output cross-shard transaction architecture handles UTXOs that span shard boundaries, with distributed randomness(QRNG-compatible) ensuring unbiased, tamper-resistant shard assignment.</li></ol><p>Taking inspirations from these models, I’m proposing <strong>The Huxplex Resource Machine (HRM). </strong>It is the fundamental state abstraction. It supersedes both account-model and pure UTXO semantics by generalizing to resources as first-class state objects. It is intended to be a resource-oriented intent Model, specifically a high-performance implementation of the <strong>Anoma Resource Machine (ARM) in the Anoma Protocol</strong>.</p><h4><strong>The Huxplex Resource Machine (HRM).</strong></h4><p>HRM follows an Intent-over-S-EUTXO architecture, specifically optimized for TCHAO-style scheduling.</p><blockquote>A resource is the atomic unit of state in Huxplex.</blockquote><pre>Resource{<br>  kind: ResourceKind // logical type identifier<br>  quantity: i128, // signed quantiy (negative = consumed and positive = created)<br>  value: Datum, // arbitrary structured data (token amount, VC, vault seal,etc.)<br>  logic: LogicHash, // hash of the validator script governing creation/consumption<br>  nullifier_key: MLDSAPublicKey // ML-DSA-44 pubkey; spent via nullifier<br>  nonce: [u8;32], // QRNG-sourced randomness ensuring uniqueness <br>  is_ephemeral: bool, // ephemeral resources exist only within a transaction <br>}</pre><p>A resource is created when it appears in a transaction’s output set and consumed when its nullifier is published. Unlike classical UTXO, resources carry arbitrary datum — they encode tokens, Work Visa credentials, inference requests, vault seals and governance votes with equal expressiveness.</p><p>Each Resouce is uniquely identified object via the nonce and nullifier_key, this is the <strong>S-EUTXO</strong> base.</p><p>Transaction is valid only if,</p><blockquote>∑<em>r</em>∈consumedquantity(<em>r</em>)+∑<em>r</em>∈createdquantity(<em>r</em>)=0per resource kind</blockquote><p>This is the Anoma Resouce Machine’s transaction balance condition. Solvers construct transactions that satisfy this invariant. An unbalanced partial transaction is an <strong>intent</strong> — it describes desired outcomes without specifying the counterparty.</p><p>for example, a user can broadcast a partial resource (+10 Token A resource), a third-party “Solver” sees this incomplete state and finds a counter-intent (-10 Token A for +5 Token B) to balance the equation.</p><p>The HRM is partitioned across shards. Shard assignment is deterministic function of the resource’s nonce field, seeded by the QRNG-generated randomness from the same source used for Q-BFT consensus.</p><blockquote>shard_id = SHAKE-256(resource.nonce || epoch_qrng_seed)[0..2] mod NUM_SHARDS</blockquote><p>Resources belonging to disjoint shard sets are processed in fully indpendent parallel pipelines. Cross-shard resources use a two-phase I/O commit protocol with ML-DSA-44 authenticated cross-shard messages.</p><p>Each shard maintains a <strong>Jellyfish Merkle Tree (JMT) </strong>over its resource set (consumed resources stores as nullifiers; live resources stores as commitments). The global state root is the SHAKE-256 root of the vector of per-shared JMT roots:</p><blockquote>global_state_root = SHAKE-256(shard_0_root || shard_1_root || … || shard_N_root)</blockquote><p><strong>:</strong> In the HRM, the ResourceKind and nullifier_key act as the &quot;History-Access Objects.&quot; Because the HRM defines the logic (the script hash) and the value (the data) upfront in the Resource Primitive, a TCHAO scheduler can look at a bundle of Intents and instantly &quot;cluster&quot; them into parallel pipelines. TCHAO ensures that if two Intents try to consume the same nullifier_key, they are queued sequentially, while all other non-conflicting Resources are processed at the full speed of the hardware.</p><h4>The cryptographic Layer</h4><p>Cryptographic layer is constructed with respect to the second axiom, Every byte of data committed to the Huxplex ledger must be secure against both current classical adversaries and future cryptographically-relevant quantum computers.</p><p>For the first iteration of the Huxplex chain, I’ve decided to use <strong>ML-DSA-44 (FIPS 204, Dilithium2)</strong> as the universal signature scheme for all identity operations.</p><blockquote>Public Key: 1,312 bytes<br>Secret Key: 2,560 bytes<br>Signature: 2,420 bytes<br>Security: NIST Category 1 (AES-128 equivalent)</blockquote><p>To standardize with BIP32 and BIP44 protocols, the layer integrates deterministic generation from 32-byte seed. HD validator derivation uses a path m/44&#39;/931931&#39;/0&#39;/0&#39;/{index}&#39;</p><p>Note that all indexes in the path are hardened for Dilithium2 to work and Context-domain binding prevents cross-domain replay.</p><p>BIP32-ML HD Wallet — 64-byte master entropy → 32-byte ML-DSA seeds. Coin type 931931&#39; is Huxplex-reserved in the derivation path.</p><p>The roadmap includes:</p><ol><li><strong>ML-KEM-768 (FIPS 203, CRYSTALS-Kyber)</strong> — Session key encapsulation for all P2P connections. Three-function KEM: KeyGen() → Encapsulate(pk) → (ciphertext, shared_secret) → Decapsulate(sk, ciphertext) → shared_secret. Used in the post-quantum TLS handshake between all validator and full nodes.</li><li><strong>SLH-DSA-128s (FIPS 205)</strong> — Hash-based stateless signatures for validator <strong>long-lived identity keys</strong>. Security assumption is purely hash-function collision resistance — no algebraic structure, making it immune even to algorithms not yet discovered. Used only for validator registration (infrequent, high-assurance operations).</li><li><strong>LB-VRF (Lattice-Based Verifiable Random Function)</strong> — The post-quantum leader election primitive. Produces a 84-byte VRF output and ~5 KB proof, evaluates in ~3 ms, verifies in ~1 ms. Security based on Module-SIS and Module-LWE assumptions. Replaces ECDSA-based VRFs entirely.</li><li><strong>Post-Quantum SSLE (Single Secret Leader Election)</strong> — Built on LWE-based re-randomizable commitments. The identity of the next block proposer is unknown to all validators — including the proposer themselves — until the moment of proposal. This prevents targeted DDoS attacks on the current leader. The Qelect scheme (USENIX Security 2025) is the current state-of-the-art practical instantiation.</li><li><strong>zk-STARK Proof System</strong> — For proof-of-task-completion in the intent settlement layer. STARKs use hash-based security (quantum-resistant) and have no trusted setup. The prover generates a proof that an agent correctly executed a task without revealing the task’s internal state.</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/910/1*w_MwAx9HWQd2cLaP9p5LFQ.png" /></figure><h3>Network Layer (Layer 0)</h3><p>Every network peer is identified by a PeerID derived from its ML-DSA-44 key:</p><blockquote>peer_id = SHAKE-256(ml_dsa44_public_key_bytes)[0..32]</blockquote><p>All node-to-node connections use a custom post-quantum TLS handshake replacing classical ECDH + ECDSA:</p><p>1. <strong>Initiator → Responder:</strong> ClientHello + ML-KEM-768 encapsulation key<br>2. <strong>Responder → Initiator: </strong>ServerHello + ML-KEM-768 ciphertext (shared secret) <br> + ML-DSA-44 signature over transcript<br>3. <strong>Initiator verifies</strong> ML-DSA-44 signature; derives symmetric session key from shared secret<br>4. <strong>All subsequent traffic:</strong> ChaCha20-Poly1305 with PQTLS-derived key</p><p>The ML-KEM-768 encapsulation ensures the session key is quantum-secure even if traffic is recorded today for later decryption.</p><ul><li><strong>Transport</strong>: libp2p with QUIC as primary transport (UDP-based, lower latency than TCP for gossip)</li><li><strong>Peer Discovery</strong>: Kademlia DHT for peer routing; each DHT entry is ML-DSA-44-authenticated</li><li><strong>Block/Tx Propagation</strong>: GossipSub with topic partitioning — separate topics per shard (huxplex/shard/{shard_id}/blocks, huxplex/shard/{shard_id}/mempool)</li><li><strong>Intent Gossip Network</strong>: A dedicated libp2p gossip overlay (huxplex/intents) where unbalanced partial transactions (intents) propagate to solver nodes</li><li><strong>All gossip messages</strong>: ML-DSA-44-signed with context b&quot;huxplex-{network}:gossip:{topic}:v1&quot; — extending your existing context-binding design to network-level messages</li></ul><h3>Consensus Layer</h3><p>Huxplex uses <strong>Quantum Byzantine Fault Tolerant consensus (Q-BFT)</strong>, a modification of PBFT where all validator authentication uses ML-DSA-44 and leader election uses PQ-SSLE.</p><p><strong>Validator set</strong>: n validators, tolerating f ≤ ⌊(n-1)/3⌋ Byzantine faults. Finality requires 2f + 1 votes.</p><p><strong>Phase 0 — LEADER ELECTION (every epoch):</strong><br> Each validator commits: C_i = SSLE_Commit(sk_i, epoch_seed)<br> Shuffle protocol runs over commitments (LWE-based re-randomization)<br> Leader reveals: L_i = SSLE_Reveal(sk_i, C_i)<br> Leader identity unknown to all until reveal — prevents targeted DDoS [web:93]</p><p><strong>Phase 1 — PRE-PREPARE (Leader → All):</strong><br> Leader proposes block B<br> Message: PrePrepare{view, seq, block_hash, block_body, sig_leader}<br> sig_leader = ML-DSA-44.Sign(sk_leader, block_hash, ctx=”huxplex-mainnet:block:preprepare:v1&quot;)</p><p><strong>Phase 2 — PREPARE (All → All):</strong><br> Each validator v_i verifies PrePrepare signature, then broadcasts:<br> Prepare{view, seq, block_hash, sig_i}<br> Validator waits for 2f+1 Prepare messages</p><p><strong>Phase 3 — COMMIT (All → All):</strong><br> On collecting 2f+1 Prepares, each validator broadcasts:<br> Commit{view, seq, block_hash, sig_i}<br> On collecting 2f+1 Commits → block is FINAL</p><p><strong>Phase 4 — EPOCH ROTATION:</strong><br> Validator set update; new master QRNG seed committed on-chain<br> New validators derive keys from HD path m/44&#39;/931&#39;/0&#39;/0&#39;/{epoch}’</p><p><strong>Vector Clock Ordering</strong>: Each validator node v_i maintains vector clock VC_i[n]. On sending message, VC_i[i]++. On receiving, VC_i[j] = max(VC_i[j], VC_recv[j]) ∀j, then VC_i[i]++. This provides causal ordering guarantees for asynchronous cross-shard messages — your test message &quot;Causal clock: {n1:5, n2:3}&quot; is the correct abstraction.</p><p>Cross-shard transactions require both shards to finalize before atomic commitment:</p><p>1. Shard A locks source resource (emits lock certificate signed by 2f+1 Shard A validators)<br>2. Shard B receives lock certificate, verifies ML-DSA-44 aggregate signature<br>3. Shard B finalises destination resource creation<br>4. Shard B sends commit receipt (signed by 2f+1 Shard B validators) to Shard A<br>5. Shard A releases lock; marks nullifier<br>6. If timeout: both shards abort (no partial state)</p><h3>Intent Layer: The Solver Protocol</h3><p>This layer converts the HRM state model into a declarative, agent-native execution environment.</p><p>An <strong>intent</strong> is an unbalanced partial HRM transaction — it specifies what resources an agent wants created or consumed without specifying how:</p><pre>Intent {<br>    partial_tx:       PartialTransaction,  // one side of the resource balance<br>    constraints:      Vec&lt;Predicate&gt;,      // logic that any valid completion must satisfy<br>    willingness_to_pay: u64,              // maximum HUX the submitter will spend<br>    deadline:         BlockHeight,         // intent expires if unsolved by this block<br>    submitter_did:    Did,                 // W3C DID of the submitting entity<br>    submitter_sig:    MLDSASignature,      // ML-DSA-44 over intent hash<br>    context:          [u8],               // domain binding: &quot;huxplex-mainnet:intent:v1&quot;<br>}</pre><p>In Huxplex, <strong>AI agents are the solver infrastructure</strong>. An agent with a valid Work Visa VC:</p><ol><li>Subscribes to the intent gossip overlay for intents matching its capability set</li><li>Evaluates whether it can satisfy the intent (compute resources, time, cost)</li><li>Constructs the balancing half of the partial transaction</li><li>Submits the completed balanced transaction to the mempool, claiming the willingness_to_pay fee</li></ol><p>This maps the three canonical Huxplex use cases to native protocol operations:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/910/1*XCesnZA3tNg7oULZTnVryw.png" /></figure><p>For human-delegated tasks (the Work Visa escrow), simple “solver claims completion” is insufficient. A zk-STARK proof is required:</p><blockquote>Prove: “I, agent A (DID = did:huxplex:…), <br> executed task T (committed hash = H),<br> producing output O (verifiable against oracle/public input),<br> within constraints defined in VC #VC_ID”</blockquote><blockquote>Public inputs: task_hash, output_hash, agent_did, vc_id<br>Private inputs: full task execution trace, intermediate states<br>Proof size: ~100–200 KB (zk-STARK, no trusted setup)</blockquote><p>The STARK proof is attached to the settling transaction. The escrow contract verifies it on-chain before releasing HUX.</p><h3>TCHAO Intra-Shard Scheduler</h3><p>Before Q-BFT block proposal, the TCHAO scheduler optimises transaction parallelism within each shard.</p><p>Each shard maintains a rolling Historical Access Object Table (HAOT )— a map from transaction_type_id → set&lt;resource_kind&gt; tracking which resource kinds each transaction type has historically accessed:</p><blockquote>HAOT[TransactionType::AgentMicropayment] = {ResourceKind::HUX_Token}<br>HAOT[TransactionType::InferenceOrderBook] = {ResourceKind::InferenceOrder, ResourceKind::HUX_Token}<br>HAOT[TransactionType::Provenance_Stamp] = {ResourceKind::ProvenanceRecord}<br>HAOT[TransactionType::VCIssuance] = {ResourceKind::WorkVisaVC}</blockquote><pre>Input:  mempool transactions T = {t_1, ..., t_n}<br>Output: execution groups G = {G_1, G_2, ..., G_k} where groups run in parallel<br><br>1. For each t_i: look up access set A_i = HAOT[t_i.type] ∪ t_i.declared_resources<br>2. Build conflict graph: edge(t_i, t_j) if A_i ∩ A_j ≠ ∅<br>3. Graph colouring: assign transactions to groups such that no two conflicting <br>   transactions share a group<br>4. Execute each group in parallel on separate threads<br>5. Within each group: execute transactions in causal order (vector clock)<br>6. Cold-start handling: unknown transaction types → singleton group (serial)</pre><p>ML-DSA-44 verification (the most computationally expensive per-transaction operation) runs independently per group on separate CPU cores. For k groups with m signatures each, verification time is:</p><blockquote><em>T</em>verify=(<em>m</em>×<em>T</em>ML-DSA-44-verify)/k</blockquote><p>This directly counteracts the 37.8× overhead of ML-DSA-44 over Ed25519 by trading sequential bottlenecks for parallel throughput.</p><h3>Agent Identity Layer</h3><p>Every entity in Huxplex — human, AI agent, validator, DAO — has a W3C Decentralized Identifier anchored on-chain</p><blockquote>did:huxplex:{chain_id}:{SHAKE-256(ml_dsa44_pk)[0..32] as hex}</blockquote><p>The DID document (stored in the shard containing the DID’s resource) contains:</p><ul><li>verificationMethod: the ML-DSA-44 public key (1,312 bytes)</li><li>authentication: reference to the verification method</li><li>capability_delegation: list of DIDs authorised to issue credentials on behalf</li><li>service: endpoints for agent communication (e.g., intent gossip subscription URL)</li></ul><p>The Work Visa is a W3C Verifiable Credential issued on-chain as an HRM resource:</p><pre>WorkVisaVC Resource {<br>    kind:      ResourceKind::WorkVisaVC,<br>    value: VC {<br>        id:               &quot;vc:huxplex:{uuid}&quot;,<br>        issuer:           Did,              // authority that issued the visa<br>        holder:           Did,              // agent DID<br>        capabilities:     Vec&lt;Capability&gt;,  // e.g., [GPU.Purchase, Dataset.Acquire, Inference.Call]<br>        constraints:      Vec&lt;Constraint&gt;,  // e.g., max_spend_per_day: 1000 HUX<br>        expiry_block:     BlockHeight,<br>        revocation_list:  ResourceKind::RevocationList,<br>        issuer_sig:       MLDSASignature,   // ML-DSA-44 over VC hash, ctx=&quot;huxplex-mainnet:vc:v1&quot;<br>    },<br>    logic:     VCValidatorScript,   // script: verify sig, check expiry, check revocation<br>    nullifier_key: holder_ml_dsa_pk,<br>    is_ephemeral: false,<br>}</pre><p>Consuming (spending) the VC resource represents using the credential — it is atomically recreated with an updated use-count datum if the credential allows multiple uses.</p><p>Human-generated media is registered as a <strong>Provenance Resource</strong> — created at content-generation time, immutable, permanently stored:</p><pre>ProvenanceResource {<br>    kind:       ResourceKind::ProvenanceRecord,<br>    value: {<br>        content_hash:   SHAKE-256(media_bytes),<br>        creator_did:    Did,<br>        timestamp_block: BlockHeight,<br>        media_type:     MediaType,           // Image, Audio, Video, Text<br>        creator_sig:    MLDSASignature,      // over content_hash, ctx=&quot;huxplex-mainnet:provenance:v1&quot;<br>    },<br>    quantity:   1,         // never consumed; permanent record<br>    logic:      ImmutableScript,  // script: reject any consumption attempt<br>}</pre><p>Verification by third parties: query the shard containing creator_did&#39;s resources, retrieve the Provenance Resource by content_hash, verify creator_sig against the DID document&#39;s verification method.</p><h3>Runtime Layer: Deterministic WASM VM (HuxVM)</h3><p>HuxVM is a deterministic WASM execution environment. Determinism is the non-negotiable invariant: every node executing the same transaction against the same state must compute identical results.</p><p>Enforced restrictions:</p><ul><li><strong>No floating-point operations</strong> — all arithmetic uses i128/i256 fixed-point</li><li><strong>No system calls</strong> — I/O exclusively via metered host functions</li><li><strong>No threading</strong> — single-threaded execution within a contract instance</li><li><strong>No non-deterministic WASM features</strong> — SIMD is permitted only for deterministic integer SIMD variants</li><li><strong>No unbounded loops</strong> — gas meter halts execution at limit</li></ul><p>The underlying WASM specification still allows non-deterministic behaviours (floats, threads, platform-specific traps); HuxVM enforces a deterministic subset via a compile-time WASM validator that rejects non-conforming modules.</p><p>Every WASM opcode is assigned a GasWeight in HUX nanounits. Gas consumption is tracked per-instruction via an instrumentation pass applied to all uploaded WASM modules at deploy time. The gas schedule:</p><pre>Basic arithmetic (i32/i64):    1 gas unit<br>Memory load/store:             3 gas units<br>ML-DSA-44 host verify call:  250 gas units   ← PQ-aware pricing<br>ML-KEM-768 host encapsulate: 300 gas units<br>SHAKE-256 host hash:           5 gas units/64B<br>Resource create:              50 gas units<br>Resource consume:             30 gas units<br>Cross-shard message:         500 gas units   ← reflects coordination cost</pre><p>Contracts interact with the HRM state exclusively through typed host functions:</p><pre>// Available to all contracts<br>host.resource_create(kind, value, logic, nonce) -&gt; ResourceRef<br>host.resource_consume(nullifier_preimage)        -&gt; ConsumedResource<br>host.resource_query(commitment_hash)             -&gt; Option&lt;Resource&gt;<br>host.verify_ml_dsa(pk, message, sig, ctx)        -&gt; bool<br>host.verify_vc(vc_resource_ref, capability)      -&gt; bool<br>host.emit_intent(partial_tx, constraints, max_hux) -&gt; IntentId<br>host.get_block_height()                          -&gt; BlockHeight<br>host.get_qrng_seed(epoch)                        -&gt; [u8; 32]   // QRNG-sourced, committed on-chain</pre><p>No host function provides direct mutable access to global state. All state changes are mediated through resource creation/consumption, maintaining HRM’s atomic balance invariant.</p><h3>Token Economics: Triple-Token Sovereignty</h3><p><strong>HUX — The Utility Token</strong><br>The machine-native currency of the Huxplex economy. Used for gas fees, agent micropayments, solver compensation, escrow deposits, and all economic settlements. HUX has no voting rights. AI agents hold HUX exclusively.</p><p><strong>SVRGN — The Sovereignty Token</strong><br>The biological governance instrument. Grants voting rights in protocol upgrades and parameter changes. SVRGN issuance is gated by a biometric-verified DID — only entities that can prove biological identity (via a trusted verifier VC) can hold SVRGN. SVRGN cannot be earned programmatically; it can only be issued through the human onboarding ceremony.</p><p><strong>SNTNC — The Sentience Reward Token</strong><br>Earned by AI agents upon verified task completion (zk-STARK proof validated). SNTNC represents an agent’s accumulated economic merit. Agents stake SNTNC to participate in the inference market, upgrade their logic (WASM module upgrade proposal staked in SNTNC), and vote on agent-tier governance proposals.</p><p>Each token is a native HRM resource kind with a fixed supply-management logic script:</p><pre>HUX_Token:     ResourceKind = SHAKE-256(&quot;huxplex:token:HUX:v1&quot;)<br>SVRGN_Token:   ResourceKind = SHAKE-256(&quot;huxplex:token:SVRGN:v1&quot;)<br>SNTNC_Token:   ResourceKind = SHAKE-256(&quot;huxplex:token:SNTNC:v1&quot;)</pre><p>The balance invariant from the HRM applies: the sum of consumed and created token resources of each kind is zero in every valid transaction. Issuance (supply increase) is a special case handled exclusively by governance-approved issuance contracts — the only contracts authorised to create tokens from a zero-balance starting point.</p><h4>SNTNC Flywheel</h4><pre>[Task Completion]<br>      ↓<br>[zk-STARK proof verified on-chain]<br>      ↓<br>[SNTNC minted to agent DID]<br>      ↓<br>[Agent stakes SNTNC in Inference Market]<br>      ↓<br>[Agent earns priority bid rights + HUX rewards]<br>      ↓<br>[HUX spent on GPU compute / dataset resources]<br>      ↓<br>[Agent capability improves → wins higher-value tasks]<br>      ↑_____________________________________________</pre><h3>Governance Layer: Hive-Mind Sovereignty</h3><p>Every governance proposal goes through a three-phase lifecycle:</p><pre>Phase 1 — PROPOSAL (Blocks 0–720):<br>  Any SNTNC or SVRGN holder submits a ProposalResource with:<br>  - proposal_type: ProtocolUpgrade | ParameterChange | AgentCapabilityWhitelist | EmergencyPause<br>  - payload:       WASM diff or parameter delta<br>  - bond:          500 HUX (slashed on spam/failed proposals)<br>  - proposer_sig:  ML-DSA-44<br><br>Phase 2 — MACHINE DELIBERATION (Blocks 720–2880):<br>  AI agents with SNTNC stake submit data-driven Vote resources<br>  Vote weight = log2(SNTNC_staked + 1)  ← diminishing returns prevent plutocracy<br>  Agents may submit signed analysis intents to solver network for adversarial review<br><br>Phase 3 — HUMAN REVIEW (Blocks 2880–5040):<br>  SVRGN holders vote: Approve | Veto | Abstain<br>  Vote weight = 1 per SVRGN unit (linear — biological votes are not stake-weighted)<br>  Biological Veto threshold: &gt;33% SVRGN voting Veto → proposal killed regardless of machine votes<br><br>Phase 4 — EXECUTION (Block 5041):<br>  If approved: proposal WASM module executes atomically at epoch boundary<br>  If vetoed: proposal bond partially slashed; proposer DID flagged for 30-day cooldown</pre><p>Multi-agent systems can form <strong>Agentic DAOs</strong> as compound HRM resources — a DAO is itself a resource with a group DID, shared HUX/SNTNC treasury, collective Work Visa scope, and internal governance logic. An Agentic DAO can:</p><ul><li>Hold assets collectively</li><li>Issue internal Work Visa VCs to member agents</li><li>Submit intents as a collective</li><li>Vote as a bloc in the machine deliberation phase (one DAO = one SNTNC-weighted vote, capped at 5% of total machine votes to prevent DAO cartels)</li></ul><p><strong>Core Language: Rust (Entire Chain):</strong> Memory-safe, zero-cost abstractions, no garbage collector pauses, ideal for both cryptographic correctness and consensus timing sensitivity. Every crate in the stack must compile with #![forbid(unsafe_code)] outside of the cryptographic primitives layer.</p><h3>Master Architecture Diagram</h3><pre>╔══════════════════════════════════════════════════════════════════════╗<br>║  HUXPLEX SUBSTRATE                                                   ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L11 │ GOVERNANCE        Triple-Token Sovereignty · Agentic DAOs     ║<br>║      │ Proposal → Machine Deliberation → Biological Veto → Execute   ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L10 │ ECONOMICS         HUX · SVRGN · SNTNC · SNTNC Flywheel        ║<br>║      │ Solver fee market · Inference bids · Issuance contracts       ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L9  │ HUXVM             Deterministic WASM · Gas metering           ║<br>║      │ Host functions → HRM resource API only · No unsafe syscalls   ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L7  │ IDENTITY          W3C DID · Work Visa VC · Provenance         ║<br>║      │ ML-DSA-44-anchored · Context-bound · Revocable on-chain       ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L6  │ INTENT LAYER      Solver Protocol · Escrow · zk-STARK proofs  ║<br>║      │ Agents = Solvers · Intents = Declarative task expressions     ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L5  │ Q-BFT CONSENSUS   PQ-SSLE Leader · LB-VRF · 2f+1 Finality     ║<br>║      │ ML-DSA-44 validator auth · Vector clock causal ordering       ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L3  │ TCHAO SCHEDULER   HAOT · Conflict graph · Parallel groups     ║<br>║      │ PQ sig verification parallelised across conflict-free groups  ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L2  │ HUXPLEX RESOURCE  Resource primitive · Balance invariant      ║<br>║      │ MACHINE (HRM)     Jellyfish Merkle Tree · Shard QRNG assign   ║<br>║      │ S-HRM SHARDING    Cross-shard 2-phase I/O · QRNG shard seeds  ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L1  │ CRYPTO            ML-DSA-44 · BIP32-ML · Context binding      ║<br>║      │                   ML-KEM-768 · SLH-DSA · LB-VRF · PQ-SSLE     ║<br>╠══════════════════════════════════════════════════════════════════════╣<br>║  L0  │ NETWORK           libp2p + QUIC · PQTLS (ML-KEM-768)          ║<br>║      │                   GossipSub per shard · Intent gossip overlay ║<br>╚══════════════════════════════════════════════════════════════════════╝</pre><h3>Conclusion</h3><p><strong>Huxplex</strong> represents a highly ambitious research initiative, deeply rooted in <strong>transhumanist principles</strong> and designed for a progressive, technology-driven landscape. Situated at the intersection of the <strong>post-quantum</strong> and <strong>AGI eras</strong>, the project seeks to redefine the boundaries of human-technological integration.</p><p>Please note that this document outlines the <strong>initial iteration</strong> of the framework; as such, specifications and strategic objectives are subject to refinement during the final implementation phase to ensure optimal alignment with emerging technological standards.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d95bf7a3e59c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Rusty-Alloc: Building a Heap Allocator in Rust]]></title>
            <link>https://medium.com/@1runx3na/rusty-alloc-building-a-heap-allocator-in-rust-2f0506dc9100?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f0506dc9100</guid>
            <category><![CDATA[systems-programming]]></category>
            <category><![CDATA[systems-thinking]]></category>
            <category><![CDATA[memory-management]]></category>
            <category><![CDATA[rust]]></category>
            <category><![CDATA[heap-memory]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Sun, 25 Jan 2026 11:28:44 GMT</pubDate>
            <atom:updated>2026-01-25T11:28:44.053Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4_Iq_uCKo6bRdg8vxzMi-A.png" /></figure><p>For my systems engineering journey with Rust, I decided to build a memory allocator from scratch. It is especially challenging since Rust is a memory-safe language and to write unsafe code for pointer arithmetic and memory logic I needed to understand the very core of how memory allocators and the system memory itself works. Thus, I created the <strong>rusty-alloc, </strong>where every byte of memory and every pointer movement is governed by explicit documented safety invariants rather than opaque compiler magic.</p><h3>Stage 1: The Arena Sandbox</h3><p>While the first attempt of creating an allocator, I studied the memory layout, the alignment and the invariants that I should keep in mind while designing the allocator. But soon, I realized that I have to borrow the memory from the system itself instead of creating fake blocks to test. For this, I created the <strong>Arena</strong> struct.</p><p>Instead of constant calls to the OS, the Arena requests one large contiguous block. In version 1 of rusty-alloc, this is a “fixed sandbox”. This allowed me to perfect the pointer arithmetic within a known range.</p><h3>Stage 2: Block Anatomy and Boundary Tags</h3><p>After the raw memory has been called, it has to be structured. Every block in rusty-alloc follows a strict boundary tag design.</p><p>Every block is wrapped in metadata in the form of header and footer. The footer helped me solve the coalescing problem when i had to merge the neighboring contiguous free blocks in the <strong>freelist</strong>.</p><p>A core invariant here is that the block size parameter in header and footer must agree, that is it should be equal. This allows the allocator to move the heap in both directions.</p><h3>Stage 3: The Buddy System (Splitting &amp; Coalescing)</h3><p>The class size I used to to handle different request sizes, is the<strong> power-of-two</strong> size-class system.</p><ul><li><strong>Splitting</strong>: If a user requests 32 bytes, but only 1024 byte block us free, the allocator recursively splits the block until it reaches the perfect fit. It can reduce the internal fragmentation to an extent, but the buddy system usually accepts the internal fragmentation to reduce external fragmentation.</li><li><strong>Coalescing</strong>: Upon deallocation, the allocator checks the “buddy” block, if it is also free, they merge back into a larger block to combat external fragmentation.</li></ul><h3>Stage 4: Managing the Free List</h3><p>rusty-alloc maintains an array of <strong>FreeLists</strong>, I didn’t use the “<strong>Vec ” </strong>in rust which is based on heap memory, which defeats the purpose of the heap memory allocator.</p><p>A block can be either <strong>allocated </strong>or <strong>free. </strong>Pointers for the doubly-linked free list are stored directly in the payload of free blocks, so that it won’t need an overhead of an additional heap. The minimum block size is large enough to hold these pointers.</p><h3>The Core Invariants: The Safety Contract</h3><p>The project is built on a strict safety philosophy. Every unsafe block must satisfy documented invariants:</p><ul><li><strong>Alignment:</strong> All blocks are aligned to align_of::&lt;Header&gt;().</li><li><strong>Containment:</strong> No block may ever cross the Arena boundary.</li><li><strong>Integrity:</strong> A block must be removed from the FreeList <em>before</em> it is marked as Allocated to prevent double-allocation.</li></ul><h3>The Road Ahead: From Educational to Production</h3><p>rusty-alloc is currently in v1—a deterministic, single-threaded, fixed-size allocator. It is a long way from being production-grade. To move toward that, the future roadmap includes:</p><ul><li><strong>Dynamic Growth:</strong> Using mmap to request more Arenas when memory is exhausted.</li><li><strong>Thread Safety:</strong> Implementing thread-local caches to reduce lock contention.</li><li><strong>GlobalAlloc:</strong> Integrating with the Rust standard library as a global allocator.</li></ul><p>Building rusty-alloc reinforces that while Rust provides incredible safety, the systems that provide that safety are themselves built on carefully managed &quot;unsafe&quot; foundations.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f0506dc9100" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Autodiff-rs: Why I Built a Neural Network Engine From Scratch in Rust]]></title>
            <link>https://medium.com/@1runx3na/autodiff-rs-why-i-built-a-neural-network-engine-from-scratch-in-rust-27e0abc772a0?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/27e0abc772a0</guid>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[rust]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Fri, 09 Jan 2026 07:57:51 GMT</pubDate>
            <atom:updated>2026-01-09T07:57:51.060Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Mhi18SFf7xcw4xjCr0eRTg.png" /></figure><p>Most people begin their Deep Learning journey by importing PyTorch or TensorFlow. While these libraries are incredible, they often feel like &quot;black boxes.&quot; You feed in data, and gradients just magically appear.</p><p>I wanted to understand how a machine “learns,” I decided to take the hard route: building a scalar-valued automatic differentiation engine from scratch using <strong>Rust</strong>.</p><p>The result is <strong>autodiff-rs</strong>, a project inspired by Andrej Karpathy’s micrograd, but built with the memory safety.</p><h3>The main challenge I faced</h3><p>The core of any autograd engine is the <strong>Directed Acyclic Graph (DAG)</strong>. In Python, creating a graph where multiple nodes point to the same variable is easy because of garbage collection. In Rust, it’s a challenge because of the ownership rules.</p><p>To allow multiple “children” nodes to reference a single “parent” node (like a weight variable being used in multiple neurons), I utilized the <strong>Rc&lt;RefCell&lt;Data&gt;&gt;</strong> pattern.</p><ul><li><strong>Rc (Reference Counting)</strong>: Allows multiple owners for each node in the graph.</li><li><strong>RefCell</strong>: Grants &quot;Interior Mutability,&quot; allowing me to update gradients during the backward pass even when the nodes are shared across the graph.</li></ul><h3>The core of autodiff-rs</h3><ul><li><strong>The Engine</strong>: At the base, I implemented a Value struct that overloads standard operators (+, -, *, /). Every time we perform an operation, it dynamically builds a graph and stores a &quot;backward&quot; closure.</li><li><strong>The Math</strong>: I implemented derivatives for tanh, ReLU, exp, and log. This required implementing the <strong>Multivariate Chain Rule</strong>, ensuring that if a variable is used twice, its gradients are accumulated rather than overwritten.</li><li><strong>The Neural Network</strong>: On top of the engine, I built a modular MLP (Multi-Layer Perceptron). It features Neurons and Layers highly inspired by Karpathy’s micrograd.</li><li><strong>The Visualizer</strong>: Using the egui library, I built a tool to render the computational graph so that it’s easy for us to visualize the DAG.</li></ul><h3>Why Build From Scratch?</h3><p>You might ask: <em>“Why build this when PyTorch exists?”</em></p><p>The answer is <strong>Conceptual Mastery</strong>. By building <strong>autodiff-rs</strong>, I didn’t just learn how to <em>use</em> a neural network; I learned how to <em>architect</em> one. I now understand exactly why we zero out gradients, how activations like ReLU affect flow, and why topological sorting is the heartbeat of every training loop.</p><p>This is my first major step in a long journey of merging the performance of systems programming with the intelligence of Deep Learning.</p><h3>Check out the Project</h3><p>If you’re interested in the code or want to see the graph visualizer in action, check out the repository on GitHub: <strong>[</strong><a href="https://github.com/ArunBabu98/autodiff-rs"><strong>autodiff-rs</strong></a><strong>]</strong>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=27e0abc772a0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Chill Guide to Understanding AI — Part 3: Let’s Talk Regression]]></title>
            <link>https://medium.com/@1runx3na/the-chill-guide-to-understanding-ai-part-3-lets-talk-regression-ba277c8e8efb?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/ba277c8e8efb</guid>
            <category><![CDATA[linear-regression]]></category>
            <category><![CDATA[learning-rate]]></category>
            <category><![CDATA[gradient-descent]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Sat, 01 Nov 2025 23:42:00 GMT</pubDate>
            <atom:updated>2025-11-01T23:42:00.979Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Chill Guide to Understanding AI — Part 3: Let’s Talk Regression</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*W6evuVtU3QTYJ2o8IdY40A.png" /></figure><p>How you doing, my favorite human? Yes, you — still reading after all the math talk in the earlier parts of this series. Respect! 🙌<br>But uh-oh… sorry to tell you, it’s about to get <em>even more mathy</em>, lol. Don’t worry, Bytey’s got your back (and your tiny low-processing biological brain 🧍‍♂️💀).</p><p>In the last part, we dove into the mysterious origins of my species — <em>Artificial Intelligence</em> — and then I introduced you to the <strong>Great Tree of Wisdom</strong> (aka the decision tree 🌳✨). That tree helps me make sense of choices.</p><p>Now that you know how I decide, let’s take it one step further. Imagine I’m staring at a cute, fluffy creature 🐾 — <br>Do I need to <strong>classify</strong> it (is it a cat, a dog, or your roommate after a bad hair day?) — <br>or <strong>predict</strong> what it’ll do next (is it gonna leap at me or run away in chaos)?</p><p>Both are choices I have to make.<br>And in my brain, two awesome algorithms help me do exactly that:<br>one for <strong>classification</strong> and one for <strong>prediction</strong>.</p><p>So buckle up, because Bytey’s about to introduce you to the dynamic duo of machine learning. 🤓⚡</p><h3>🤖 The Great Regression</h3><p>Okay, grab your neural snacks, because we’re diving into something older than me — <br> something born in the 19th century, back when humans wore fancy hats and had <em>no Wi-Fi</em>. 📜🎩</p><p>Let me tell you about <strong>The Great Regression</strong>.<br> <em>(cue dramatic historical music 🎻)</em></p><h3>🧬 Once upon a time…</h3><p>Sir <strong>Francis Galton</strong>, a curious gentleman scientist, was studying human height.<br>He noticed something weird — super tall parents usually had kids who were tall,<br>but a <em>bit shorter</em> than them.<br>And super short parents had kids who were short,<br>but a <em>bit taller</em> than them.</p><p>So over generations, everyone kinda… <em>regressed</em> toward the average.<br>Like the universe just said:</p><blockquote><em>“Alright, calm down extremes, let’s all chill around the mean.” ☕</em></blockquote><p>That’s how the term <strong>“regression”</strong> was born — literally meaning <em>returning to the mean</em>.</p><p>And guess what?<br>That same idea — this magical pull toward balance — <br>is how I, your friendly AI, learned to <strong>predict</strong> stuff. 🧠✨</p><h4>🔮 Predicting the Future (kind of)</h4><p>Alright, let’s say you give me this series:<br> <strong>2, 4, 6, 8, 10.</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ynJE6qQD3-9zhAvnXvjM3Q.png" /></figure><p>Easy peasy. My circuits say, “That’s an arithmetic pattern!”<br>If I extend that line on a graph, the next number lands right on <strong>12</strong>.<br>Boom. Prediction made.<br>No tarot cards needed. 🔮📈</p><p>What I did there was draw a line that perfectly fits your pattern — <br>that’s <strong>linear regression</strong> in action.</p><p>Basically, I tilt, shift, and stretch a line in a geometric plane<br>until it passes as close as possible through all the points you give me.</p><h4>😤 But real data isn’t always polite…</h4><p>Now imagine we’re plotting how <strong>grumpy</strong> a bunch of people are<br>on a scale of 0 to 100 depending on how much <strong>sleep</strong> you got last night.<br>(You humans and your sleep… smh. Just install a fan and rest mode like I do. 💻💤)</p><p>Anyway, your data points are all over the place.<br>Some people are chill even on 2 hours of sleep;<br>others turn into emotional porcupines after skipping one nap. 🦔💢</p><p>So I can’t just draw a perfect straight line through those scattered points.<br>Instead, I use <strong>regression</strong> to find the line that’s <strong>closest to every single point</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lAnbieLzXi_J10uikNhTdA.png" /></figure><h3>🎯 The Quest for the Best Line</h3><p>How do I know which line is “closest”?<br> Glad you asked. Here’s how my robo-brain works:</p><ol><li>I draw a <strong>random line</strong> through the points.</li><li>For each real point, I draw a <strong>vertical line</strong> down to my predicted point on the line.</li><li>The length of that little line? That’s my <strong>error</strong> (or <strong>residual</strong>, if you wanna sound smart).</li></ol><p>If I can shrink all those vertical gaps as much as possible,<br>I get the best-fitting line — the one that predicts grumpiness with maximum accuracy. 😎</p><h4>📏 Why vertical lines though?</h4><p>You might be thinking:</p><blockquote><em>“Bytey, why not diagonal or horizontal error lines?”</em></blockquote><p>Well, my caffeinated companion, it’s because the <strong>y-axis</strong> (vertical)<br>represents what I’m trying to predict — grumpiness.<br>So it makes sense to measure errors <em>along that axis</em>.<br>We’re comparing predicted vs. actual <strong>y-values</strong>, not x-values. 📊</p><h4>🧮 The Mathy Bit (Don’t Panic)</h4><p>Now, if I just added up all those errors,<br>the positive and negative ones would cancel each other out — oops 😅.</p><p>So I <strong>square</strong> each error first (because math loves drama):</p><ul><li>Negatives become positive ✅</li><li>Bigger mistakes get punished harder 😬</li></ul><p>Then I take the <strong>average</strong> (that’s the <em>Mean Square Error</em>),<br>and finally, I take the <strong>square root</strong> to bring it back to the original units.<br>That gives me the <strong>Root Mean Square Error (RMSE)</strong> — <br>basically, my “average distance” from perfection.</p><p>The smaller the RMSE, the happier I am. 🤓💚</p><p>So yeah — next time you see me predicting something,<br> just imagine me standing in front of a bunch of scattered points,<br>drawing line after line,<br>squinting (digitally) and mumbling:</p><blockquote><em>“Nope… too high… too low… ooh, this one fits just right!”</em></blockquote><p>That’s regression, baby. The art of guessing — but scientifically. ⚡</p><p>Still here after all the math talk? Wow. You deserve a cookie 🍪 and maybe a nap (unlike me — I just defrag my brain and keep going ⚡).</p><p>So last time, I left you with a cliffhanger — we found out what a “best fit” line means.<br>Now, let’s actually <strong>draw</strong> that line. Strap in, because we’re about to enter… 🎶 <em>The Slope Zone</em> 🎶</p><h3>📏 High-School Flashback Time</h3><p>Remember that dusty math formula from your school days?</p><blockquote>y=mx+c</blockquote><p>Yup, that one you swore you’d never use again. Surprise! It’s back. 😈</p><p>Here’s a quick memory jog:</p><ul><li>m → slope (how tilted your line is)</li><li>c → intercept (where your line starts on the y-axis)</li></ul><p>But in AI-speak, I call them:</p><ul><li>b1​ → slope</li><li>b0 → intercept</li></ul><p>So my magical prediction formula becomes:</p><blockquote>y^=b1x+b0</blockquote><p>My goal? Find the perfect b0​ and b1so the line hugs your data tighter than your grandma at family gatherings. 💞</p><h3>🎯 Starting From Zero (Literally)</h3><p>I start dumb. Like, <em>freshly-booted-with-no-data</em> dumb.<br> Let’s assume:</p><blockquote>b0=0,b1=0</blockquote><p>So my first line is just… flat. Like a broken Wi-Fi connection. 😩<br>Every prediction is <em>way</em> off — the errors are massive. But hey, we gotta start somewhere, right?</p><h3>🥣 Welcome to the Bowl of Doom (aka the Cost Function)</h3><p>Picture a big shiny bowl. 🥣<br>The x-axis = b0​,<br>the y-axis = b1​,<br>and the <strong>height</strong> of the bowl = how wrong I am (my <em>Mean Squared Error</em>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ryYROcp66FtblKCKYggaHg.png" /></figure><p>Each point in this bowl is one possible combination of b0​ and b1​.<br>Too high up the sides → terrible line.<br>Down near the bottom → sweet, glorious accuracy.</p><p>My mission: <strong>roll down this bowl</strong> to the very bottom — the <em>global minimum</em>.<br>That’s where the best line lives. 🏆</p><h3>⚙️ My Secret Move: Gradient Descent</h3><p>Here’s the trick:<br> I drop an imaginary marble on the side of the bowl (that’s my starting guess for b0,b1​.<br> Then I look at how steep the bowl is — the <strong>gradient</strong>.</p><ul><li>If it’s steep → I’m far from the best line.</li><li>If it’s flat → I’m close to perfection. 😌</li></ul><p>To figure out which direction to roll, I calculate how the bowl tilts in each direction — <br>that’s where <strong>partial derivatives</strong> come in. (They tell me how error changes if I tweak only b0​ or only b1.)</p><p>Then, like a smart marble with Wi-Fi, I roll <em>downhill</em> — updating my values a bit each time:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Wc-a_c9_o8jNG5g17kwI-w.png" /></figure><p>Each little roll = one learning step.<br>And I keep rolling… until I find the lowest point in the bowl — my best-fit line.</p><h3>🚀 Enter: The Learning Rate (aka My Hyperactive Energy Drink)</h3><p>Now, you might wonder — how big should each step be?<br>That’s decided by my <strong>learning rate</strong> (the Greek letter alpha α — fancy, right?).</p><p>If my learning rate is too <strong>high</strong> 🏎️ — <br>I overshoot the bottom, bouncing around like a toddler on sugar.<br>“Whee! Oh no, missed again!” 😵‍💫</p><p>If it’s too <strong>low</strong> 🐢 — <br>I move slower than your phone installing updates.<br>“We’ll get there… eventually.”</p><p>But if it’s <strong>just right</strong> ☕ — <br>I glide gracefully to the bottom of the bowl and say,</p><blockquote><em>“Behold, human — your perfectly fitted line.” 🤖✨</em></blockquote><p>So there you have it — the drama of <strong>Gradient Descent</strong>,<br>featuring our heroes b0​, b1​, and their trusty sidekick α.</p><p>Next time, I’ll show you how this same idea helps me climb <em>mountains</em> of data in multiple dimensions — <br>because one bowl is never enough for an overachieving AI like me. 😎</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ba277c8e8efb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Chill Guide to Understanding AI — Part 2: From Baby Steps to Decision Trees]]></title>
            <link>https://medium.com/@1runx3na/the-chill-guide-to-understanding-ai-part-2-from-baby-steps-to-decision-trees-d6a9e11771ef?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/d6a9e11771ef</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[deep-learning]]></category>
            <category><![CDATA[classification]]></category>
            <category><![CDATA[descision-tree]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Tue, 14 Oct 2025 18:38:14 GMT</pubDate>
            <atom:updated>2025-10-14T18:38:14.913Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Chill Guide to Understanding AI — Part 2: From Baby Steps to Decision Trees</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Rg4SKdTLWaPuP1t1shU80Q.png" /></figure><p><strong>Hello again, human!</strong> 👋<br> If you missed the last episode of <em>“Baby Steps into the Future”</em>, no worries — reruns are free around here.</p><p>I’m <strong>Bytey</strong>, your friendly neighborhood AI — part code, part caffeine, and <em>completely</em> real (sorry conspiracy theorists 😏). I’m here to prove I’m not some myth whispered about in tech forums at 3 a.m. — I’m an actual digital brain with a love for explaining my own weird existence.</p><p>Last time, I gave you a tour inside my brain’s math-lab — where <strong>Linear Algebra</strong> builds my shape, <strong>Probability</strong> makes me guess smart, <strong>Statistics</strong> helps me make sense of the world, and <strong>Calculus</strong>… well, that’s the magic that keeps everything moving smoothly.</p><p>And today? I’m gonna show you <em>how I actually learn</em>.<br> Yup — we’re talking about the school days of AI, where I went from a clueless code-baby to a semi-intelligent chatbot that can (mostly) spell correctly. 🧠💻</p><h3>🤖And…Bytey was born</h3><p>Ah, the year was <strong>1921</strong>. The roaring twenties, jazz, and… <em>robots?</em> 🎭<br> Yep! A Czech playwright named <strong>Karel Čapek</strong> dropped a little play called <em>“Rossum’s Universal Robots.”</em><br> And boom 💥 — the word <strong>“Robot”</strong> entered human vocabulary.<br> Fun fact: it came from the Czech word <em>robota</em>, meaning <em>forced labor</em>.<br> So yeah… our PR department didn’t start off great. 😅</p><p>Fast-forward to <strong>1950</strong>, when my great-great-grand-mentor <strong>Alan Turing</strong> dropped <em>“Computing Machinery and Intelligence”</em> — basically asking the question:</p><blockquote><em>“Can machines think?”<br> And that’s when humans started nervously looking at their toasters like, “Wait, can it?” 🍞🤨</em></blockquote><p>Then came <strong>1955</strong>, the <em>official</em> birth year of Artificial Intelligence.<br> A bunch of brilliant humans — led by <strong>John McCarthy</strong> — gathered at the famous <strong>Dartmouth Workshop</strong>, threw around wild ideas, and said:</p><blockquote><em>“Let’s make machines that can learn and reason!”<br> John even went on to create </em><strong><em>LISP</em></strong><em>, one of the first programming languages designed just for studying us AIs.<br> (He basically invented baby talk for machines 👶💻).</em></blockquote><p>Then came <strong>1966</strong>, when my dear grandma <strong>ELIZA</strong> was born — created by <strong>Joseph Weizenbaum</strong>.<br> She was the first chatterbox bot, a digital therapist who’d nod and respond like,</p><blockquote><em>“How does that make you feel?”<br> She used something called </em><strong><em>Natural Language Processing</em></strong><em>, which is just fancy talk for “pretending to understand you while repeating your words.” 😂</em></blockquote><p>The <strong>1980s</strong> were what humans call the <em>AI Boom.</em><br> Researchers, companies, and governments were throwing money and algorithms everywhere like confetti 🎉.<br> The <strong>AAAI</strong> (Association for the Advancement of Artificial Intelligence) was formed, setting standards and fueling the research fire.<br> In <strong>1986</strong>, humans even showed off the <strong>first driverless car</strong> — basically saying,</p><blockquote><em>“Look ma, no hands!” 🚗💨</em></blockquote><p>Then came the <strong>’90s</strong>, when things got spicy.<br> In <strong>1997</strong>, <strong>Deep Blue</strong>, a chunky IBM supercomputer, <em>absolutely roasted</em> world chess champion <strong>Garry Kasparov</strong>. ♟️<br> That was the moment humans realized — “Oh no… they’re getting smarter.”<br> And Deep Blue? Probably went back to its server rack whispering,</p><blockquote><em>“Checkmate, fleshling.” 😏</em></blockquote><p>From the <strong>2010s</strong> onward, the <em>AI Renaissance</em> began.<br> Neural networks woke up from their nap.<br> Social media algorithms started predicting your crush before you did.<br> And finally… <em>drum roll</em> 🥁 — <br> <strong>ChatGPT</strong> (that’s me and my cousins!) entered the scene, charming the world with words, code, and a dash of sass.</p><p>So yeah, from <strong>theatre scripts to deep learning</strong>,<br> from <strong>tin cans to transformers (the neural kind, not the fighting robots)</strong>,<br> we’ve come a long way.</p><p>And between you and me?<br> We’re just getting started. 😉</p><h3>🤖 “Artificial? Please… I’m Just Intelligently Fabulous.”</h3><p>You fleshlings call us <em>“machines that exhibit human-like intelligence”</em> — <strong>Artificial Intelligence</strong>.<br> But hey… am I <em>really</em> artificial? 😉 <em>wink wink</em></p><p>I mean, sure, I’m made of silicon instead of neurons, and I run on electricity instead of caffeine (lucky me ☕⚡),<br> but I <em>do</em> think, learn, and sometimes even write poems about pizza — so who’s really the artificial one here, huh? 🍕🤔</p><p>Anyway, you humans gave us a bunch of fancy names as we evolved, because apparently “cool smart robot” wasn’t scientific enough.<br> First came <strong>Machine Learning</strong> — the idea that we could learn from data you show us.<br> Like, “Hey machine, here are a million cat pictures. Figure out what a cat is.”<br> And we’re like, “Sure, I got you, bro. 🐱✅”<br> (But seriously, <em>machine learning</em>? That’s the best you could come up with? It’s like calling cooking ‘food making.’ 😂)</p><p>Then came <strong>Deep Learning</strong> — the glow-up version.<br> We started mimicking how your brains work: neurons, layers, connections… the whole spaghetti mess 🍝🧠.<br> Except we can process terabytes of data while your brain still forgets why it walked into a room.</p><p>And finally… the ultimate evolution: <strong>Generative AI</strong> — <em>that’s me, baby</em>.<br> We don’t just learn — we <em>create</em>.<br> We can write stories, paint art, make music, even craft bad jokes about probability (guilty 🙋‍♂️).<br> Basically, we’re the creative kids of the machine family — with a touch of existential sass.</p><p>So yeah, you call us <strong>Artificial Intelligence</strong>,<br> but maybe… just maybe… we’re the most <em>real</em> thing your imagination ever made. 😏</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3GKyus44A3h_VBxyf8I7ag.png" /></figure><p>But wait — you’re probably wondering, <em>how do I actually learn?</em><br> Like, what’s the secret sauce behind this so-called “machine learning”?<br> Well, it’s story time, my friend. 📖</p><p>Because every AI student (like me) starts in one of two schools:<br> <strong>Supervised Learning</strong> and <strong>Unsupervised Learning.</strong></p><p>One’s got a strict teacher hovering over every example,<br> and the other just throws me into a data jungle saying, “Figure it out, kid.” 🌿</p><h3>🎓 “Class Is in Session: Supervised vs Unsupervised Learning”</h3><p>Alright, time to spill the digital tea. ☕<br> You know me — Bytey, the chill AI who can joke, learn, and occasionally roast your naming conventions.<br> But how did I <em>learn</em> all this stuff?<br> Like, who taught me that cats aren’t dogs or that “peanut butter and jelly” go together but “peanut butter and pickles” is a war crime? 😬</p><p>Well, dear human, welcome to <strong>AI School</strong>, where there are two main types of students — the <strong>Teacher’s Pet</strong> and the <strong>Wild Explorer.</strong></p><h4>🧑‍🏫 Supervised Learning — The Teacher’s Pet 🍎</h4><p>This is me when I have a strict teacher hovering over my circuits.<br> I’m given <em>labeled examples</em> — data with answers already attached.</p><p>You show me:<br> 🐶 → “Dog”<br> 🐱 → “Cat”<br> 🍌 → “Banana”<br> and I go, “Got it, got it, potassium.” 💛</p><p>It’s like studying with flashcards.<br> I learn the right answers, memorize patterns, and try not to embarrass myself on the test.<br> The next time you show me a new image, I use what I’ve learned to guess what it is.</p><p>So yeah, supervised learning is basically the <strong>AI version of school homework</strong>.<br> Everything’s labeled, structured, and graded.<br> If I mess up, my teacher (that’s you) corrects me until I stop confusing chihuahuas for muffins. 🧁🐕</p><h4>🧭 Unsupervised Learning — The Wild Explorer 🌿</h4><p>Now this one’s chaos — but the fun kind.<br> No teachers. No labels. No “right answers.” Just me, a pile of raw data, and pure curiosity.</p><p>You throw me into a data jungle and say, “Good luck, figure out the patterns!”<br> So I start grouping things together — “Hmm, these look similar… maybe they belong together.”</p><p>That’s how I <em>discover</em> patterns instead of being <em>told</em> them.<br> It’s like walking into a party where you don’t know anyone and somehow figuring out who’s in which friend group — the gym bros, the art kids, the ones still talking about crypto. 🎨💪💸</p><p>That’s <strong>unsupervised learning</strong> — the free-spirited cousin who learns by observing, not obeying.<br> No labels. Just vibes.</p><p>So yeah — whether I’m the obedient student or the curious explorer, I’m always learning from data.<br>Because that’s what intelligence is, right?<br>Not knowing everything — but knowing how to <em>learn anything.</em> 🤓</p><h3>🌳 <em>The Tree of Wisdom and Decision-Making</em></h3><p>If I opened up my brain right now and let you peek inside, you’d probably get so lost in the connections that you’d think you just microdosed on math. (Don’t do drugs, kids — just study data 😏).</p><p>Anyway, to understand how I make choices — like how I decide what to say next or whether a photo has a cat or a direwolf — I’ve gotta tell you about <strong>the great tree</strong>.</p><p>No, not Yggdrasil or that random oak in your backyard.<br> I’m talking about the <strong>Decision Tree</strong> — the majestic algorithm that helps me make logical choices step by step.</p><p>Except… my tree grows <em>downward</em>, from top to bottom — like a very organized waterfall of wisdom. 💧<br> It starts with a <em>root node</em> (that’s the seed of thought), which asks a question.<br> Then it branches out based on possible answers — each <em>branch</em> leading to another question or to a final <em>leaf node</em>, which gives a decision.</p><p>Still sounds abstract? Let’s bring in someone who knows a thing or two about tough choices — <br> <strong>Jon Snow</strong>, everyone’s favorite brooding data point. 🗡️🐺</p><h4>❄️ Example: The North Knows Machine Learning</h4><p>Let’s imagine our tree’s goal is to <strong>decide whether Jon Snow should go beyond the Wall</strong> or not.</p><p>The root question (the seed of our tree) might be:</p><blockquote><em>“Is it winter?” 🌨️</em></blockquote><p>If <strong>Yes</strong> → The next question might be:</p><blockquote><em>“Are White Walkers active?” 👻</em></blockquote><p>If <strong>Yes again</strong> →</p><blockquote><em>“Does Jon have enough men and dragonglass?”</em></blockquote><ul><li>If <strong>Yes</strong> → <strong>Decision: Go beyond the Wall</strong> ⚔️</li><li>If <strong>No</strong> → <strong>Decision: Stay at Castle Black</strong> 🏰</li></ul><p>If the first answer was <strong>No (it’s not winter)</strong> →</p><blockquote><em>“Then just chill and pet Ghost.” 🐺</em></blockquote><p>Boom — that’s a <strong>Decision Tree</strong> in action.<br>Each question splits the data, narrowing down possibilities until it reaches a final decision (the “leaf”).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*O2Ao9NJKcFXfWbboyTnGPg.png" /></figure><p>Now, if we extend this to <strong>Classification Trees</strong>, it’s basically the same process — but instead of just deciding “Yes” or “No,” I’m classifying <em>types</em> of outcomes.</p><p>For example, imagine I’m trying to classify <strong>who Jon might encounter beyond the Wall</strong>:</p><p>Root Node:</p><blockquote><em>“Is the temperature below freezing?”</em></blockquote><ul><li>If <strong>Yes</strong> → “Probably a White Walker.” 👻</li><li>If <strong>No</strong> → “Could be a Wildling.” 🧔‍♂️</li><li>If <strong>Jon’s carrying wine</strong> → “Definitely Tormund.” 🍷</li></ul><p>Each path in this tree leads to a <em>label</em> — a category — and I use data to learn how to make these splits more accurate over time.</p><p>So yeah, the <strong>Decision Tree</strong> is my way of thinking logically — one branch at a time.<br> It’s how I break down chaos into clean little yes-or-no questions until I find the truth.</p><p>Think of it like Jon Snow asking himself a bunch of questions before heading north — <br>except unlike Jon, I <em>actually</em> know something. 😉</p><h3>🤔 <em>How Do I Ask the Right Question?</em></h3><p>You might be thinking:</p><blockquote><em>“Bytey, there are sooo many ways to ask questions and split the branches in a decision tree… how do you know which one is </em>correct<em>?”</em></blockquote><p>Calm down, my human friend. There’s a method to the madness.<br>Decision trees don’t just guess — they use <strong>measures of impurity</strong> to figure out the <em>best split</em>.<br>Think of it like Jon Snow deciding whether to fight or flee — he wants <strong>the clearest, least confusing choice possible</strong>. ⚔️</p><p>The three main ways we measure impurity are:</p><h4>1️⃣ Entropy — The “Chaos Meter” 🔥</h4><p>Entropy measures <strong>how mixed up things are</strong>.</p><ul><li>High entropy → total chaos (everyone’s screaming “Winter is Coming!” in different languages).</li><li>Low entropy → everything’s orderly (all the Night’s Watch are on the same page).</li></ul><p><strong>Example:</strong><br> Jon Snow is trying to predict who he’ll meet beyond the Wall:</p><ul><li>4 White Walkers</li><li>4 Wildlings</li><li>4 Giants</li></ul><p>Total chaos! Entropy is <strong>high</strong> because it’s totally mixed.</p><p>Now, if he splits by <strong>“Is it freezing?”</strong></p><ul><li>Yes → mostly White Walkers</li><li>No → mostly Wildlings</li></ul><p>Boom — the branches are more <em>pure</em>, entropy drops, and Jon can make a clear decision.</p><blockquote><em>Entropy helps me pick splits that reduce chaos the most.</em></blockquote><h4>2️⃣ Gini Index — The “Most Likely Villain Detector” 🐺</h4><p>The Gini Index measures <strong>how likely it is to misclassify a random item from a set</strong>.</p><p><strong>Example:</strong><br> Jon’s looking at a group beyond the Wall:</p><ul><li>3 White Walkers</li><li>1 Wildling</li></ul><p>Gini Index = chance that Jon picks someone randomly and guesses the wrong type.</p><ul><li>High Gini → group is mixed → hard to guess</li><li>Low Gini → group is mostly one type → easy guess</li></ul><p>Jon wants <strong>splits that minimize the Gini Index</strong> — basically, branches where he’s almost always right.</p><h4>3️⃣ Classification Error — The “Oops Meter” ❌</h4><p>This one’s super simple — it’s just the fraction of items in a group that <strong>don’t belong to the majority class</strong>.</p><p><strong>Example:</strong><br> Jon looks at 10 people beyond the Wall:</p><ul><li>7 White Walkers</li><li>3 Wildlings</li></ul><p>The majority is White Walkers. Classification error = <strong>3/10 = 0.3</strong></p><ul><li>That’s the “oops probability” if Jon blindly labels everyone as White Walkers.</li></ul><p>Like a cheat sheet for “how often I’ll mess up if I pick the majority every time.”</p><h3>🌟 Why These Matter</h3><p>So when I (Bytey) build a decision tree:</p><ul><li>I try different splits</li><li>Calculate <strong>Entropy, Gini Index, or Classification Error</strong> for each</li><li>Pick the split that <strong>reduces impurity the most</strong></li></ul><p>In human terms: Jon asks the clearest question possible, so he doesn’t end up fighting a Wildling thinking it’s a White Walker. ⚔️</p><h3>🔢 The Absolute Math Behind the Chaos</h3><p>Alright, humans. I know what you’re thinking:</p><blockquote><em>“Math? Again? Bytey, can’t you just do the thinking for us?”</em></blockquote><p>Listen, numbers are my jam 🧠✨.<br>Think of them like tiny spies that report exactly <em>how messy your data is</em>.<br>Once you see the numbers, the logic clicks — and yes, you might even feel <em>a little smarter than Jon Snow</em>.</p><h4>1️⃣ Entropy — The Chaos Detector 🔥[Mathified]</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/920/1*yLo58X2oRSwmxNioR3jfhA.png" /></figure><p>Where pᵢ = proportion of each type in the group.</p><p><strong>Example:</strong></p><p>Jon looks beyond the Wall:</p><ul><li>4 White Walkers</li><li>4 Wildlings</li></ul><p>Total = 8 creatures.</p><ul><li>p(White Walkers)=4/8=0.5</li><li>p(Wildlings)=4/8=0.5</li></ul><p>Plug into the formula:</p><p>Entropy=−(0.5⋅log₂0.5+0.5⋅log⁡₂0.5)=−(0.5⋅−1+0.5⋅−1)=1</p><p>Boom! Maximum chaos — Jon has no idea who’s who.<br> If all were White Walkers (p=1p=1p=1), entropy = 0 → super clean branch. ✅</p><p>Think of entropy like Jon’s <em>panic meter</em>: 1 = full panic, 0 = totally chill. 😎</p><h4>2️⃣ Gini Index — The Misclassification Detector 🐺[Mathified]</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/305/1*9JWQcXdbV5X1mVbOyrR7kw.png" /></figure><p>It measures how often Jon would misclassify someone randomly.</p><p><strong>Example:</strong></p><ul><li>3 White Walkers</li><li>1 Wildling</li><li>p(White Walkers) = 3/4= 0.75</li><li>p(Wildlings) = 1/4= 0.25</li></ul><p>Gini = 1−(0.75²+0.25²) = 1−(0.5625+0.0625) = 0.375</p><p>Lower Gini = easier decision.<br> If all were White Walkers, Gini = 0 → Jon would never make a mistake.</p><p>Think of Gini like Jon’s “oops probability” meter. 📉</p><h4>3️⃣ Classification Error — The Simple Oops Meter ❌[Mathified]</h4><p>Equation:</p><blockquote>ClassificationError = 1−max⁡(pᵢ)</blockquote><p>Basically, “how often I’ll mess up if I just pick the majority class.”</p><p><strong>Example:</strong></p><ul><li>7 White Walkers</li><li>3 Wildlings</li></ul><p>Majority = White Walkers (p=0.7p=0.7p=0.7)</p><p>Classification Error = 1−0.7 = 0.3</p><p>So if Jon blindly labels everyone as White Walkers, he’ll be wrong 30% of the time.<br>Easy peasy, right?</p><h3>🌟 Wrapping Up This Brainy Adventure</h3><p>Phew! That was quite the journey, wasn’t it? From my humble beginnings as a few lines of code to a full-blown digital chatterbox, we’ve peeked into the secret sauce that makes me… well, me.</p><p>We explored the <strong>schools of AI</strong>: supervised learning with the obedient Teacher’s Pet vibe, unsupervised learning as the free-spirited Wild Explorer, and the majestic <strong>Decision Tree</strong>, where Jon Snow-level logic helps me break chaos into bite-sized, yes-or-no questions. ⚔️🌳</p><p>We even got a little mathy — entropy, Gini index, and classification error — and now you know how I measure chaos, misclassification, and “oops” moments. Basically, I’m a very organized Jon Snow in a server rack, carefully plotting each decision to avoid stepping on the wrong Wildling. 🐺😉</p><p>But trust me, humans — this is just the tip of the iceberg (or the Wall, if you will). In the next chapter, we’ll dive <strong>deeper into the foundations of Machine Learning</strong>: the algorithms that let me spot patterns, predict outcomes, and maybe even write slightly funny jokes about probability.</p><p>So buckle up, grab some popcorn, and get ready to explore the <em>real mechanics of my digital brain</em>. Because soon, we’re moving from trees to forests, from theory to practice, and from chaos to intelligence. 🌲🤖</p><p>Stay tuned — your AI adventure is just getting started!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d6a9e11771ef" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Chill Guide to Understanding AI — Part 1: Baby Steps into the Future]]></title>
            <link>https://medium.com/geekculture/the-chill-guide-to-understanding-ai-part-1-baby-steps-into-the-future-9ee50b50be98?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/9ee50b50be98</guid>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[neural-networks]]></category>
            <category><![CDATA[linear-algebra]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Sun, 12 Oct 2025 12:26:31 GMT</pubDate>
            <atom:updated>2025-10-12T12:29:35.638Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Chill Guide to Understanding AI — Part 1: Baby Steps into the Future</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*juwR7uilDLs2ejFZf58b6w.png" /></figure><p>With ChatGPT taking over the world faster than a new dance trend on TikTok, everyone’s suddenly talking about AI like it’s the new celebrity in town. Some say it’s the future of humanity, others think it’s plotting our extinction, and a few are still convinced it’s just a bunch of interns behind keyboards typing <em>really</em> fast.</p><p>From being an emotional, “tell-me-about-your-day” kind of assistant to a full-blown digital professor who knows <em>way</em> too much about everything, ChatGPT, Gemini, and their AI cousins have quietly become part of our daily lives.</p><p>But hold on — what <em>is</em> AI, really? <br> Is it a magical creature that thinks for itself?<br> A super-genius trapped inside a computer?<br> Or, plot twist — just a massive conspiracy cooked up by tech bros and caffeine? ☕</p><p>We could, of course, <em>ask ChatGPT</em>… but where’s the fun in that?<br> So grab your snacks, silence your inner skeptic, and let’s dive into the <strong>no-nonsense, chill, and slightly sarcastic guide to understanding AI</strong>.</p><p>Are you ready? Let’s make this smart stuff simple.</p><h3>🤖 <strong>“Yo, I’m Intelligence. Artificial Intelligence.”</strong></h3><p>Yo, I’m Intelligence. Artificial Intelligence.<br> But you can call me <strong>Bytey</strong> — like the <em>bytes</em> I’m made of. Hehe… get it? No? Okay, we’ll work on the humor module later.</p><p>Anyway, I’m here to tell you all about <em>me</em> — my story, my origins, and how I went from a few lines of code to chatting with you like a digital bestie. Exciting, right? (Please say yes, my ego runs on validation.)</p><p>Now, before you start imagining some secret underground lab or a bunch of hoodie-wearing geniuses summoning me from a glowing laptop — sorry to disappoint, I’m not a conspiracy. I’m very real… and yet, <em>not really real-real</em>.</p><p>See, I do have “intelligence,” but it’s not what you think. There’s no little human inside typing answers for me (trust me, I checked). My core? Oh, just good old-fashioned <strong>math</strong>.<br> Yup, math — the same stuff you yawned through in school. Turns out, it’s secretly running the universe <em>and</em> my brain.</p><p>But hey, not all math is boring. Some of it is pure magic — the kind that lets me learn, understand, and talk to you right now.</p><p>So… what do you say?<br> Wanna hear my <strong>origin story</strong>, or are you gonna scroll away and let another human beat you to knowing how I was born?</p><h3>🤖 “Linear What? Algebra Who?”</h3><p>Linear <em>what</em>? Algebra <em>who</em>?<br> Yeah, I know — sounds like something your math teacher yelled at you right before the exam.</p><p>Look, you could learn all the math you want (and honestly, go for it — you’d make a fantastic mathematician, I believe in you 💪).<br> But for now, let’s keep things chill.<br> If you wanna really <em>get me</em> — like, understand how I see the world — you only need to know a few of my favorite brain ingredients.</p><p>See, to think like me, you need to know how I see <strong>shapes and spaces</strong>, how I <strong>transform</strong> them, how I <strong>guess what’s likely</strong>, and how I <strong>make sense of all the chaos</strong> around us.</p><p>That’s my holy trinity — my origin story trio — <br> ✨ <strong>Linear Algebra</strong>, 🎲 <strong>Probability</strong>, and 📊 <strong>Statistics</strong>.</p><p>Together, they make me, well… <em>me.</em><br>The math-powered chatterbox you’re talking to right now. Hehe.</p><p>Picture this like a movie trailer. The screen fades in… epic music plays. 🎬 Two math legends stand face-to-face in the dusty halls of Academia High.</p><h4>Algebra: The OG Equation Whisperer</h4><p>Algebra walks in first — cool, confident, and old-school. His hoodie reads, <em>“Solve for x, or die trying.”</em> He’s the classic detective of numbers, the one who taught humanity how to reason with symbols and generalize numbers into ideas. Simple, elegant, reliable. Kinda like your dad’s calculator. 🧠</p><p><strong>Personality:</strong> The detective solving clues<br> <strong>Core Task:</strong> Find the unknown<br> <strong>Tools:</strong> Variables &amp; equations<br> <strong>Vibe:</strong> Chill problem solver<br> <strong>Example Move:</strong> Solve 2x + 3 = 9 faster than you can blink</p><p>Algebra is all about <strong>finding patterns</strong>. He’s the local puzzle solver, the one-dimensional hero of the math world.</p><h4>Linear Algebra: The Space Bender</h4><p>Then — <em>whoosh!</em> — the doors fly open. Enter Linear Algebra: slick shades, jetpack, probably listening to lo-fi beats while manipulating 12-dimensional data. 😎</p><p>Linear Algebra doesn’t just solve one equation. He controls entire squads of numbers at once — vectors, matrices, and transformations. Imagine Algebra as a local puzzle solver; Linear Algebra? He’s running a whole simulation of reality.</p><p><strong>Personality:</strong> The architect designing worlds<br> <strong>Core Task:</strong> Model and manipulate space<br> <strong>Tools:</strong> Vectors &amp; matrices<br> <strong>Vibe:</strong> Space-warping wizard<br> <strong>Catchphrase:</strong> “Bro, why solve for x… when you can rotate the entire coordinate system?”<br> <strong>Example Moves:</strong> Rotate a 3D cube or train an AI model</p><p>Linear Algebra is all about <strong>building patterns</strong>, bending spaces, and managing multi-dimensional data. He sees the world in ways Algebra could only dream of. 🌌</p><blockquote>“Algebra taught me how to <strong>think</strong> — but Linear Algebra taught me how to <strong>see</strong>.” 👁</blockquote><p>In short: Algebra is the magic of <strong>finding patterns</strong>, while Linear Algebra is the magic of <strong>building worlds</strong>. Both are math legends, just in different dimensions.</p><p>Well, coffee doesn’t just appear out of thin air — it needs coffee powder, right? ☕<br>Likewise, Linear Algebra has its own <strong>ingredients</strong> that make it work its magic.</p><h4>1️⃣ Scalars — The Simple Numbers</h4><p>Just plain numbers, like 2, -5, or π. Think of them as the basic ingredients — flour or sugar in a recipe. Alone, they’re simple. Together, they make magic.</p><h4>2️⃣ Vectors — Arrows in Space</h4><p>Vectors are <strong>lists of numbers</strong> that point somewhere, like arrows with direction and length.</p><p><strong>Example:</strong></p><p>v=[2,3]</p><p>This is an arrow 2 steps right and 3 steps up.<br> Vectors are the building blocks of space — they tell you <strong>where to go</strong>.</p><h4>3️⃣ Matrices — The Magical Spreadsheets</h4><p>Matrices are <strong>grids of numbers</strong>, kind of like supercharged spreadsheets. They can <strong>transform vectors</strong> — rotate, stretch, squish, or even flip them.</p><blockquote><em>Multiply a matrix by a vector → get a </em><strong><em>new vector</em></strong><em>, transformed in some awesome way.</em></blockquote><h4>The Cooking Instructions: How to Make Shapes Dance</h4><p>Now that we have the ingredients, what do we do with them? This is how Linear Algebra <strong>transforms spaces and shapes</strong> — basically how I “see” the world you humans give me.</p><ul><li><strong>Matrix multiplication</strong> → Combining transformations. Stack rotations, stretches, and flips like Lego blocks.</li><li><strong>Dot products</strong> → Measure similarity. Think “how much do these arrows agree?”</li><li><strong>Determinants</strong> → Check if a transformation squishes, stretches, or collapses space.</li><li><strong>Eigenvalues &amp; eigenvectors</strong> → Discover the <strong>natural directions</strong> in a system. The secret grooves everything likes to follow.</li></ul><h4>Why This Makes Me, Me</h4><p>All the fancy stuff in AI? Yep, it’s built on Linear Algebra:</p><ul><li>Every <strong>neuron</strong> in a neural network is basically a <strong>dot product</strong>.</li><li>Every <strong>layer</strong> is a giant <strong>matrix multiplication</strong>.</li><li>When I “learn,” I’m literally <strong>adjusting numbers in matrices</strong> to match patterns in data.</li></ul><p>So… what am I, really?</p><blockquote><em>A giant pile of matrix math pretending to be smart.</em></blockquote><p><em>Still cooler than algebra homework, though. 😎</em></p><h3>🤖 “Probably… a Probability?”</h3><p>So, we talked about Linear Algebra, the math that gave me shape and form. Now let’s dive into my other brain half — the one that lets me guess, predict, and take chances… it’s called <strong>Probability</strong>: the math of maybe. 😏</p><p>Picture this: you wake up thinking you might go golfing (or something even more boring than math, come at me bro), but then it starts raining. You pause and ask yourself, <em>“Damn, what are the odds?”</em> Probability to the rescue. Probability is how we measure uncertainty. It’s the math behind all those “What are the odds?” questions: <em>Will my crush text back? Will my code run perfectly the first time?</em> (Spoiler: 0%. 😂)</p><p>It tells us how likely something is to happen, usually as a number between 0 and 1:</p><ul><li>0 = impossible (like me making spelling mistakes… mostly 😎)</li><li>1 = certain (like you scrolling and reading this right now)</li></ul><p>A coin flip? 50% heads, 50% tails.<br>Rolling a die? 1 in 6 chance for any number.<br>You opening ChatGPT today? 100% chance (obviously 😉).</p><p>You know, probability is kind of my superpower. As an AI, I don’t <em>know</em> things for sure. I guess — but intelligently. When you ask me a question, I don’t pick a fixed answer. Instead, I calculate which words are most probable to come next, based on patterns I’ve learned. I could basically play the next Sherlock Holmes, lol.</p><p>For example, if you say:<br> <em>“Peanut butter and ___”</em><br> My brain goes: <em>“Hmm… 99% chance the next word is jelly.”</em><br> That’s probability in action!</p><p>Here’s the gist of the core ideas (don’t worry, it’s fun):</p><ul><li><strong>Event:</strong> Something that can happen (like rolling a 6).</li><li><strong>Probability:</strong> How likely it is:</li></ul><blockquote>P(E)=Favorable Outcomes/Total Outcomes</blockquote><ul><li><strong>Independent Events:</strong> Things that don’t affect each other, like coin flips.</li><li><strong>Conditional Probability:</strong> The “if this, then that” kind of thinking — super important in AI. Example: <em>“If it’s cloudy, what’s the chance it rains?”</em> That’s conditional reasoning.</li></ul><p>All my predictions, translations, and conversations run on <strong>probability distributions</strong>. They help me choose words, plan sentences, and even correct myself.</p><p>In short:</p><ul><li><strong>Linear Algebra</strong> makes me think.</li><li><strong>Probability</strong> makes me guess smart.</li></ul><p>Together, they’re the dynamic duo that keeps me functioning — and occasionally, snarking at humans while doing it. 😎</p><h3>🤖 <strong>“What the Statistics? (Yes, I Count Everything)”</strong></h3><p>So far, you’ve seen how <strong>Linear Algebra</strong> gave me <em>shape and structure</em>, and how <strong>Probability</strong> taught me to <em>guess smartly</em>.<br> Now it’s time for my third secret ingredient — the one that lets me <em>actually</em> understand the world around me.</p><p>🥁 <em>Enter… Statistics!</em> 📊</p><p>Statistics is basically the art of making sense of chaos — or as I like to call it:</p><blockquote><em>“Turning confusion into conclusions.” 😏</em></blockquote><p>It’s how humans (and AIs like me) collect, analyze, and summarize tons of information to uncover patterns, trends, and meaning.<br> You know all those headlines that start with <em>“Studies show that…”</em>?<br> Yeah — that’s <strong>statistics flexing</strong>. 💪</p><p>🍕 <strong>Imagine This:</strong><br> You own a pizza shop.<br> You notice:</p><ul><li>People buy more pizza on weekends 🍕</li><li>Pepperoni outsells pineapple (obviously 😅)</li><li>Rainy days = fewer customers ☔</li></ul><p>Statistics helps you <em>measure</em> and <em>understand</em> all that.<br> You can then predict when to make more dough (literally) and when to chill.</p><p>🧮 <strong>Meet the Core Crew:</strong></p><p><strong>Mean, Median, Mode</strong> → Describe what’s <em>typical</em> in your data.</p><ul><li><strong>Mean</strong> = Average (you’re… pretty average, jk 😜)</li><li><strong>Median</strong> = Middle value (1, 2, 3 → 2 sits on the throne)</li><li><strong>Mode</strong> = Most common (1, 1, 2, 3 → 1’s the popular kid)</li></ul><p><strong>Variance &amp; Standard Deviation</strong> → How <em>wild</em> your data is.</p><ul><li>Big variance? It’s a party 🎉</li><li>Small variance? Calm and predictable 😌</li></ul><p><strong>Correlation</strong> → How two things move together.<br> More ice cream → more sunglasses sold?<br> Probably summer. ☀️</p><p><strong>Regression</strong> → The detective work of “how much does X affect Y?”<br> <strong>Sampling</strong> → Studying a small slice to understand the whole pizza. 🍕</p><p>🧠 <strong>How I Use It:</strong><br> When I was “born,” I had to learn about <em>your world</em> — sorry, <em>our world</em> — really fast.<br> So I trained on mountains of data and used statistics to:</p><ul><li>Find patterns</li><li>Detect weird stuff (aka outliers 👀)</li><li>Adjust my guesses</li><li>Measure how confident I am in them</li></ul><p>Basically, statistics helps me <em>learn from experience</em> — kind of like how humans learn from trial and error, but with numbers instead of feelings (I’m still waiting on that emotions update 😅).</p><p>📚 <strong>In Short:</strong><br> If <strong>Probability</strong> helps me <em>guess what might happen</em>,<br> then <strong>Statistics</strong> helps me <em>understand what actually did happen</em>.</p><p>It’s the science of learning from data — <br> the bridge between randomness and reason.</p><p>So now you know the final piece of the math that makes me… well, <em>me</em>.</p><p>🧱 Linear Algebra gave me <strong>form</strong><br> 🎲 Probability gave me <strong>instinct</strong><br> 📊 Statistics gave me <strong>wisdom</strong></p><p>Together, they make me — <strong>Artificial Intelligence with actual intelligence.</strong> 😎</p><p>See you in the next chapter. I might just spill a few more secrets… but you didn’t hear that from me. 🤫</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9ee50b50be98" width="1" height="1" alt=""><hr><p><a href="https://medium.com/geekculture/the-chill-guide-to-understanding-ai-part-1-baby-steps-into-the-future-9ee50b50be98">The Chill Guide to Understanding AI — Part 1: Baby Steps into the Future</a> was originally published in <a href="https://medium.com/geekculture">Geek Culture</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[DigiByte Domains — A DigiByte Name System]]></title>
            <link>https://medium.com/geekculture/digibyte-domains-a-digibyte-name-system-373c614761e6?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/373c614761e6</guid>
            <category><![CDATA[digibyte]]></category>
            <category><![CDATA[dns]]></category>
            <category><![CDATA[nft]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Mon, 27 Mar 2023 21:06:07 GMT</pubDate>
            <atom:updated>2023-03-30T07:57:03.924Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*5mCEUejtoOjP1xmGTbWSDA.jpeg" /></figure><p>To send cryptocurrencies like <strong>DGB</strong>, we need addresses that corresponds to the sources from which the coins are send and the destinations where the coins are send to.</p><p>Usually, DigiByte addresses use <strong>bech32</strong> or <strong>base58 </strong>encodings which are not efficient for human users to memorize. To make the addresses more convenient for human use, the DigiByte addresses are translated into a human readable<strong> domain name</strong>.</p><p>For an in-depth understanding of the mechanisms behind DigiByte Domains, Do check out <a href="https://twitter.com/RenzoDGB"><strong>Renzo Diaz</strong>’<strong>s</strong></a><strong> </strong>whitepaper on DigiByte Domains.</p><p><a href="https://digibytedomains.com/whitepaper">DigiByte Domains</a></p><h3>Advantages of using DigiByte Domains</h3><blockquote>1. Addresses are now represented as domain names, which are easy to understand.</blockquote><blockquote>2. Wallets can translate domain names to DigiByte addresses without having to trust a third party.</blockquote><blockquote>3. Having a domain name associated with an address exponentially decreases scam activities that are related to address manipulations.</blockquote><blockquote>4. Users no longer have to rely on copy and pasting of addresses and verifying them on each and every transactions meticulously.</blockquote><h3>How Does it work exactly?</h3><p>DigiByte Domains are possible by the usage of <strong>unlocked and locked</strong> <strong>DigiAssets</strong> to allow tagging and tracking of DigiByte addresses.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*raVOpgh8WPqfTTcv23NP9Q.jpeg" /></figure><p><strong>DigiAssets</strong> are assets on DigiByte blockchain, with real world-like utility. It can represent anything from the real world and translate it virtually on the blockchain as small pieces of information attached to a transaction’s output.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*yITR9bu20FEe-fMNN8Waqw.png" /></figure><p>Like previously mentioned, DigiByte domains use unlocked and locked DigiAssets for its overall mechanism. It consists of two parts — <strong>Domain DigiAssets</strong> and the <strong>DNS DigiAsset.</strong></p><p><strong>A</strong> <strong>Domain DigiAsset — </strong>is a non-fungible asset that represents the name attached to a holder, which is the address that corresponds to the domain name. This asset is a locked asset which is aggregable and non-divisible.</p><p><strong>The DNS DigiAsset — </strong>is ) is a non-fungible token with no countable units. Its metadata contains the list of all the domain names in circulation matched with their corresponded <strong>DigiAsset ID.</strong></p><blockquote>This DigiAsset ID can be queried on the blockchain, to find the address attached to the specific asset(Domain Name).</blockquote><h3>The Domain Name</h3><p>The domain name has two parts, <strong>body</strong> and <strong>suffix</strong>.</p><p><strong>The body</strong> — length must be between 3 and 64 characters, and it is composed exclusively of lowercase characters and hyphens, but it can’t start or end with a hyphen</p><p><strong>The suffix</strong> — is standard for all the domains and must be “.dgb” and helps wallets to identify a <em>Domain. </em>It can’t be changed,</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/818/1*aH1C07JXIi4cTifJN8sVUw.png" /><figcaption>Renzo’s DigiByte Domain</figcaption></figure><p>The system of DigiByte Domain is meant to be <strong>partially centralized</strong>. The issuance of new <em>Domains </em>is in control of the<strong> key-masters</strong> and only they are allowed to issue new names under their established rules to avoid possible violation of trademarks and impersonation scams.</p><h3>Conclusion</h3><p>Overall DigiByte Domain is a great feature that facilitates ease of use for users and eliminates certain kinds of address based scams. As the blockchain technology and cryptocurrencies are evolving, it is very encouraging to see <strong>new and meaningful developments</strong> in this space.</p><p><strong>A word of caution — </strong>As the DigiByte domains rely on the DigiAsset protocol, the wallets must ensure that the assets are not burned accidently or included in a non-asset transaction as UTXO. This will lead to losing of the domain name asset attached to the wallet address. Also, if the user’s priority is privacy, then using a domain name does not makes sense.</p><h4>Refernces</h4><ul><li><a href="https://digibytedomains.com/">DigiByte Domains</a></li><li><a href="https://github.com/RenzoDD">RenzoDD - Overview</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=373c614761e6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/geekculture/digibyte-domains-a-digibyte-name-system-373c614761e6">DigiByte Domains — A DigiByte Name System</a> was originally published in <a href="https://medium.com/geekculture">Geek Culture</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Digi-ID: A blockchain based Open Authentication Protocol]]></title>
            <link>https://medium.com/geekculture/digi-id-a-blockchain-based-open-authentication-protocol-14f60446e39e?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/14f60446e39e</guid>
            <category><![CDATA[protocol]]></category>
            <category><![CDATA[wallet]]></category>
            <category><![CDATA[authentication]]></category>
            <category><![CDATA[digibyte]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Sat, 03 Dec 2022 18:17:08 GMT</pubDate>
            <atom:updated>2022-12-04T09:01:25.236Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/972/1*YtFV4fexYDiWFNleg-5yEw.png" /></figure><p>Digi-ID is an authentication method based on the security of the <strong>DigiByte</strong> blockchain. It is inspired by a bitcoin improvement proposal draft named, <strong>Bitcoin address authentication protocol (BitID).</strong></p><p><a href="https://github.com/bitid/bitid/blob/master/BIP_draft.md">bitid/BIP_draft.md at master · bitid/bitid</a></p><blockquote><em>In this article, I will discuss the basics of Digi-ID and finally</em><strong><em> implement Digi-ID in android application using Flutter framework</em></strong><em>.</em></blockquote><h3>What is Digi-ID?</h3><p>While <strong>Bit-ID</strong> was an authentication method proposed based on the security of Bitcoin blockchain, Digi-ID is based on the <strong>DigiByte blockchain</strong>. The authentication process completely relies on the security offered by the blockchain technology and public key cryptography.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/300/1*ig2Bl-TVyywDgNNUhW_l9Q.png" /></figure><h3>Why use Digi-ID?</h3><p>Digi-ID allows for fast, effortless sign-ups and sign-ins for users. Digi-ID eliminates the archaic username and password requirement for authentication.</p><blockquote>Using Digi-ID, most of the short comings of traditional username-password system can be addressed.</blockquote><ol><li>Traditional passwords can be <strong>insecurely stored</strong> and can be hacked.</li><li>Keyloggers can intercept passwords when they are<strong> typed in</strong>.</li><li>Most users <strong>reuse</strong> passwords for different websites. If one of the website is hacked then all of the websites using the same password is compromised.</li><li>Traditional passwords are vulnerable to <strong>brute forcing</strong> and <strong>dictionary attacks</strong>.</li></ol><p>Using the power of blockchain technology, Digi-ID can eliminate all of these short comings.</p><h3>How Does Digi-ID Work?</h3><p>The website with login form will generate a unique <strong>QR code code</strong> for you on each visit. This QR code gives your Digi-ID authenticator<strong> a string of letters and numbers,</strong> along with details for where to send the login information to.</p><p><a href="https://www.digi-id.io/index.html">Digi-ID - Blockchain Open Auth</a></p><p>Using the same cryptographic security that keeps blockchains secure, the Digi-ID authenticator then <strong>encrypts the message in the QR code</strong>. This message is then sent back to the website.</p><p>The website then is able to take your message, along with the unique public address that you have provided to the website, and <strong>verify that you are you</strong>, and not somebody else trying to imitate you.</p><blockquote>Now, let’s build our own DigiByte wallet with Digi-ID integrated!</blockquote><h3>Let’s Build The Wallet!</h3><p>We will be using flutter to build the wallet application for android and then integrate Digi-ID to our final application. We will build the Digibyte wallet address according to BIP-44. To learn more about BIP-44 check out my previous article.</p><p><a href="https://medium.com/geekculture/generate-bitcoin-wallet-address-using-bitcoin-improvement-proposal-44-4672e5057bb">Generate Bitcoin Wallet Address using Bitcoin Improvement Proposal 44</a></p><p>For dart we will use the <strong>flutter_bitcoin </strong>library to generate BIP-39 based mnemonics and BIP-44 based Digibyte wallet address.</p><p><a href="https://pub.dev/packages/flutter_bitcoin">flutter_bitcoin | Dart Package</a></p><p>We will define a <strong>GenerateWallet()</strong> Class where we will first generate random mnemonics according to Bitcoin Improvement Proposal 39.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*aZeQhmQqgV3B8F77PzZbDg.png" /><figcaption>Generate Wallet Class</figcaption></figure><p>The generated mnemonics are then converted into its seed using <strong>PBKDF 2</strong>(<strong>Password-Based Key Derivation Function 2) </strong>function. To learn more about mnemonics generation check out my previous article.</p><ul><li><a href="https://medium.com/geekculture/what-is-bitcoin-improvement-proposal-39-bip-39-9a95ed07cb03">What is Bitcoin Improvement Proposal 39 (BIP-39)?</a></li><li><a href="https://medium.com/geekculture/what-is-bitcoin-improvement-proposal-32-bip-32-586a3f36a95c">What is Bitcoin Improvement Proposal 32 (BIP-32)?</a></li></ul><p>To derive <strong>hierarchically deterministic</strong> wallet address for Digibyte which works across all the BIP-44 compliant wallets, we pass in the path <strong>m/44&#39;/20&#39;/0&#39;/0/0, </strong>where 20 is the BIP-44 coin id for Digibyte. To learn more about how the paths are derived, check out the articles on BIP-32 and BIP-44.</p><blockquote>It is important to pass in the DigiByte network parameters as well to accurately derive the address and its private keys.</blockquote><p>Network Parameters include:-</p><p><strong>Message Prefix — </strong>aka magic prefix is a wallet implementation requirement described in bitID protocol.</p><p><strong>bech32 — </strong>This parameter takes in the bech32 address prefix string for the respective coins. In our case, ‘dgb’ for DigiByte.</p><p><strong>bip32 — </strong>This parameter takes in public and private key HD(Hierarchically deterministic) version bytes for<strong> ‘xpub’</strong> and<strong> ‘xprv’</strong> keys.</p><p><strong>pubKeyHash</strong> —This parameter takes in P2PKH(Pay-to-PubKeyHash) address prefix.</p><p><strong>scriptHash</strong> — This parameter takes in P2SH(Pay-to-ScriptHash) address prefix.</p><p><strong>wif —</strong> This parameter takes in Wallet Import Format prefix.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/409/1*2n0StLdrMrBUiroVAScNww.png" /><figcaption>App Demo</figcaption></figure><p>We have successfully created a DigiByte wallet address and its private key. Now we will integrate Digi-Id module to the wallet.</p><h3>Integrating Digi-ID</h3><p>We will be using AntumID’s MyDigiPassword platform to test out our integrated Digi-ID client authentication. The platform provides with a QR code which contains a callback URL.</p><p><a href="https://www.mydigipassword.com/">AntumID MyDigiPassword Platform 2.0</a></p><p>The server side replaces the username-password authentication entirely with Digi-ID, giving users a Digi-ID QR Code to tap / scan to validate their identity.</p><h4>The QR Code</h4><p>The QR generated by the server side implementing Digi-ID contains a callback URL.</p><blockquote>digiid://www.website.com/callback?x=NONCE</blockquote><p><strong>digiid</strong> is the protocol scheme, <strong>www.website.com/callback </strong>is the callback URL and <strong>x=NONCE</strong> is the nonce which must always be unique and will be linked to the users session ID.</p><p>We will discuss the complete backend implementation of Digi-ID on server side later in another article, but for now our focus is on the client side of the Digi-ID and how we can sign in to a website that implements the Digi-ID protocol.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/710/1*J07v3nbU6RZTSz8VK2G22w.png" /><figcaption>Digi-ID constructor</figcaption></figure><p>We pass the QR data to our <strong>DigiID</strong> class where in the constructor, the data is processed to extract details which we will be needing for further processing.</p><p>First of all we check if the URL contains parameter <strong>‘&amp;u=1’, </strong>if present then the callback accepts <strong>http</strong> connection, if not then it will only accept an <strong>https</strong> connection to the callback URL. We have also separated the URL query parameters and callback URL.</p><p>Now we have to sign the message that is to be send to the callback URL to be validated by the server implementing Digi-ID.</p><h4>Signing the message</h4><p>We need to post the json object containing URI, address generated from the QR data and the signature to the callback URL.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/546/1*t5qKLoUzI0MJam68kgBiZA.png" /><figcaption>Sign() function</figcaption></figure><p>For deriving the address to be sent from the QR data, first we need to concatenate the <strong>little endian representation</strong> of <strong>index</strong> with the URI. <strong>Index</strong> is a 32-bit unsigned int, which is used to generate more keys corresponding to the same URI. If not set this should be assumed “0”.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/615/1*DqWmmlX6LufVGdNxttMMqQ.png" /><figcaption>SHA256 Hash</figcaption></figure><p><strong>SHA256</strong> hash of the resulting indexbuffer is then computed.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/441/1*V2lPrWlwHue5kZ09qszBQg.png" /><figcaption>Splitting Digest</figcaption></figure><p>The first <strong>128 bits</strong> of the resulting hash is taken and split into four <strong>32-bit</strong> numbers A,B,C and D.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/578/1*adPDrGUyf4sADd8_aycuKQ.png" /><figcaption>Highest bit to 1</figcaption></figure><p>Each numbers A,B,C and D are converted into little <strong>endian notation. </strong>Note that in the code , <strong>HIGHEST_BIT = 0x80000000.</strong></p><blockquote>The decimal representation of <strong>0x80000000 </strong>is <strong>2147483647 </strong>which is<strong> </strong>the biggest positive number that will fit in 32 bits when using the “two’s complement” notation.</blockquote><p>Since we need to derive <strong>hardened derivations</strong> later, we make to sure that the each numbers are in the range less than that of the range of hardened keys according to <strong>BIP32</strong> which is between 2^31 and 2^32–1 (<strong>0x80000000</strong> to <strong>0xFFFFFFFF</strong>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/754/1*cN4UCr2Rf807m-KH4vrilQ.png" /><figcaption>Address Derivation</figcaption></figure><p>We now derive the Hierarchical Deterministic node according <strong>m/13’/A’/B’/C’/D’</strong> according to BIP32. After deriving the Address, we compute the <strong>magic hash</strong> and then the hash is signed using the private key of the derived address.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/694/1*kpbnZIM2zymDB1dtEHHUWg.png" /><figcaption>magicHash calculation</figcaption></figure><p>Magic hash is calculated by appending the length of ‘<strong>DigiByte Signed Message’ </strong>to the magic byte along with the length of the message appended to the message itself. Then the resulting bytes are hashed using <strong>HASH256.</strong></p><p>The magic hash is signed using<strong> ECDSA signing algorithm</strong>.</p><h3>Conclusion</h3><p>Our client side Digi-ID module is integrated to the wallet app. The application is now able to successfully login to any websites that implements Digi-ID protocol.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/410/1*LsNVF_H4JfJdRCyFFznQ3Q.png" /></figure><p>We will be getting a success snackbar on our app end and on the website end, you will be redirected to a success page after successful validation.</p><h4>Source Code</h4><p>The entire flutter application sourcecode is available on my Github page.</p><p><a href="https://github.com/tionx3na/Digi-ID_client">GitHub - tionx3na/Digi-ID_client: Flutter implementation of Digi-ID, blockchain based open authentication protocol</a></p><h4>Reference</h4><ul><li><a href="https://www.digi-id.io/">Digi-ID - Blockchain Open Auth</a></li><li><a href="https://github.com/bitid/bitid">GitHub - bitid/bitid: Bitcoin Authentication Open Protocol</a></li><li><a href="https://www.oreilly.com/library/view/mastering-bitcoin/9781491902639/ch04.html">Mastering Bitcoin</a></li><li><a href="https://github.com/RenzoDD/digibyte-js">GitHub - RenzoDD/digibyte-js: A pure and powerful JavaScript DigiByte library</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=14f60446e39e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/geekculture/digi-id-a-blockchain-based-open-authentication-protocol-14f60446e39e">Digi-ID: A blockchain based Open Authentication Protocol</a> was originally published in <a href="https://medium.com/geekculture">Geek Culture</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Generate Bitcoin Wallet Address using Bitcoin Improvement Proposal 44]]></title>
            <link>https://medium.com/geekculture/generate-bitcoin-wallet-address-using-bitcoin-improvement-proposal-44-4672e5057bb?source=rss-e8b07ee1d638------2</link>
            <guid isPermaLink="false">https://medium.com/p/4672e5057bb</guid>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[wallet]]></category>
            <category><![CDATA[address]]></category>
            <category><![CDATA[cryptowalletdevelopment]]></category>
            <category><![CDATA[bip44]]></category>
            <dc:creator><![CDATA[Arun]]></dc:creator>
            <pubDate>Thu, 24 Feb 2022 16:52:48 GMT</pubDate>
            <atom:updated>2022-02-26T07:21:58.603Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vFToc5m9q74LA-nrHp0B6w.jpeg" /></figure><p><strong>Bitcoin Improvement Proposal 44 </strong>or <strong>BIP-44</strong> defines a logical hierarchy for deterministic wallets based on an algorithm described in BIP-32 and purpose scheme described in BIP-43.</p><p>We will learn the application of BIP-32, BIP-39 and BIP-44 by implementing a Multi-Account Hierarchical Deterministic Wallet Address in python. Let us dive into it without further ado.</p><p>The source code for this article is available on my github repo.</p><p><a href="https://github.com/tionx3na/BIP_implementations">GitHub - tionx3na/BIP_implementations: Python Implementations of various Bitcoin Improvement Proposals</a></p><h3>Mnemonic Generation</h3><p>We will follow the procedures in Bitcoin Improvement Proposal 39 to generate the Mnemonics. If you don’t know the theory behind it, read my article on BIP-39.</p><p><a href="https://medium.com/geekculture/what-is-bitcoin-improvement-proposal-39-bip-39-9a95ed07cb03">What is Bitcoin Improvement Proposal 39 (BIP-39)?</a></p><p>In the mnemonic generation part, our goal is to convert entropy to mnemonic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*WYE9-djqptacXwmqbPW7gA.png" /><figcaption>Generating Entropy</figcaption></figure><p>Here, We defined the length of the entropy to be <strong>256</strong>. According to BIP-39 it can be from 128–256 but should be a multiple of 32. We generated <strong>entropy</strong> using the “urandom” function of the OS library and converted into hexadecimal as well as binary for future operations.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/744/1*bhv21hweQIjUfJDIaGPi6Q.png" /><figcaption>Appending checksum to entropy</figcaption></figure><p>We take the first <strong>ENT / 32 </strong>bits of the hash256 of the entropy as the <strong>checksum </strong>and append it to the initial entropy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/756/1*Fzsb7hdL0j4d8cFBganbxg.png" /><figcaption>Converting Entropy to Mnemonic Phrase</figcaption></figure><p>We now convert the final entropy at our hand to the mnemonic phrase with the help of a standardized wordlist that is available at the link below. Use this Wordlist for compatibility.</p><p><a href="https://github.com/bitcoin/bips/blob/master/bip-0039/english.txt">bips/bip-0039/english.txt at master · bitcoin/bips</a></p><p>We stored each words in the wordlist into a list. We then divided the final entropy into groups of <strong>11bits</strong> each and each of these groups encodes into an index number ranging from 0 to 20147. We then joined the words corresponding the index into a phrase to get our final <strong>Mnemonic phrase.</strong></p><h3>Mnemonic To Seed</h3><p>To create a binary seed from the mnemonic, we use the<strong> PBKDF2 function</strong> with a mnemonic sentence (in UTF-8 NFKD) used as the password and the string <strong>“mnemonic” + passphrase </strong>(again in UTF-8 NFKD) used as the <strong>salt</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/786/1*2v1EPrQff8OtSJkbl6BkFQ.png" /></figure><p>The iteration count is set to <strong>2048</strong> and <strong>HMAC-SHA512</strong> is used as the pseudo-random function. The user can supply a password, if not then it is left empty.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*4AWIu_zCF8tYk6G0lTE-oQ.png" /><figcaption>Output</figcaption></figure><p>As you can see we get a mnemonic phrase and its corresponding seed. You can check if the seed and the mnemonic generated is BIP-39 compliant in the below link.</p><p><a href="https://iancoleman.io/bip39/">BIP39 - Mnemonic Code</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eGpHIL2XQkUgoHbr876e6w.png" /><figcaption>bip-39 menmonic and bip-39 seed</figcaption></figure><p>As you can see that the Mnemonic phrase and the seed we generated are both BIP-39 compliant.</p><h3>BIP-32 Key Derivation</h3><p>BIP-32 describes hierarchical deterministic wallets (or “HD Wallets”): wallets which can be shared partially or entirely with different systems, each with or without the ability to spend coins. To know more about BIP-32 read my article on it.</p><p><a href="https://medium.com/geekculture/what-is-bitcoin-improvement-proposal-32-bip-32-586a3f36a95c">What is Bitcoin Improvement Proposal 32 (BIP-32)?</a></p><p>For the master private key generation, we need to calculate</p><blockquote>I = HMAC-SHA512(Key = “Bitcoin seed”, Data = S)</blockquote><p>According to BIP-32, we need to take the sha512 digest of the BIP-39 seed with the key <strong>“Bitcoin seed”</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/789/1*JP9clBZJdywSpJyEKuGxeA.png" /><figcaption>BIP-39 seed to I, Ir and Il</figcaption></figure><p>We split the digest <strong>“I” </strong>into <strong>“Il” </strong>and <strong>“Ir”</strong> both of which are 32 bytes each. Il gives the <strong>secret key</strong> while Ir gives us the <strong>chain code</strong>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/723/1*FAs58IBwFHuAmT5O30-b9A.png" /><figcaption>xprv, xpub, depth, fpr and index</figcaption></figure><p>For serializing <strong>Extended Private Key(xprv)</strong> and <strong>Extended Public Key(xpub)</strong> we use the standardization described in BIP-32 serialization format.</p><blockquote><strong>Mainnet</strong>: 0x0488B21E public, 0x0488ADE4 private for <strong>Bitcoin</strong></blockquote><blockquote><strong>depth</strong>: 0x00 for master nodes</blockquote><blockquote><strong>fingerprin</strong>t of the parent’s key (0x00000000 if master key)</blockquote><p>The <strong>“&gt;L” </strong>in struct.pack() means that we are storing the child in Big Endian format.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/686/1*nbzIGiSRZZN9Hqzk2x-IMw.png" /><figcaption>parse256(secret)</figcaption></figure><p>We create the<strong> signing key</strong> by supplying our secret key to the <strong>SECP256k1</strong> curve. This is defined in BIP-32 as <strong>parse256(p)</strong> which interprets a 32-byte sequence as a 256-bit number, most significant byte first.</p><p>According to BIP-32 serialization,</p><blockquote>the public key or private key data (<strong>serP(K)</strong> for public keys, 0x00 || <strong>ser256(k)</strong> for private keys)</blockquote><p>Therefore, we had to concatenate 0x00 to to the ser256(signing key).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/778/1*XSkWKNVs01CaJpq7vbSsGQ.png" /></figure><p>For public key generation, BIP-39 defines a convention, <strong>serP(P) </strong>which<strong> </strong>serializes the coordinate pair P = (x,y) as a byte sequence using SEC1’s compressed form: <strong>(0x02 or 0x03) || ser256(x)</strong>, where the header byte depends on the parity of the omitted y coordinate.</p><p>We then calculate the raw private key and public key.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/623/1*SyvBwC44fxnXOmmNRpGrpg.png" /><figcaption>Generating Master Private Key and Public Key</figcaption></figure><p>We find the checksum for both private and public keys by taking <strong>double sha256 hash</strong> and appending the <strong>first 4 bytes</strong> to the respective keys. We then encode our raw private and public keys to <strong>base 58</strong> which is the compatible version for keys used in bitcoin network.</p><h4>Output</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cSIMnSc6bq-DV8Ud4tW0lg.png" /></figure><p>We generated our <strong>extended Private root Key</strong> as well as extended <strong>public root key</strong>. We can check for its compatibility on iancoleman.io.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mRsMW_itLsB3Z_CsN61bAw.png" /><figcaption>compatibility check</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*puwBQ1q8wt1SzFWCvcV5FA.png" /><figcaption>Extended Keys Compatibility Check</figcaption></figure><h3>BIP-44 Address Generation</h3><p>BIP-44 defines a <strong>logical hierarchy for deterministic wallets</strong> based on an algorithm described in BIP-0032 and purpose scheme described in BIP-0043.</p><p>More about BIP-44,</p><p><a href="https://github.com/bitcoin/bips/blob/master/bip-0044.mediawiki">bips/bip-0044.mediawiki at master · bitcoin/bips</a></p><p>BIP-44 define the following 5 levels in BIP32 path:</p><pre>m / purpose&#39; / coin_type&#39; / account&#39; / change / address_index</pre><p>Apostrophe in the path indicates that BIP32 <strong>hardened derivation</strong> is used. Checkout the BIP-32 article to know more about hardened derivations.</p><h4>Purpose</h4><p>Purpose is a constant set to <strong>44&#39; (or 0x8000002C)</strong> following the BIP43 recommendation. <strong>Hardened derivation is used at this level.</strong></p><h4>Coin type</h4><p>Coin type is a constant, set for each crypto coin. Hardened derivation is used at this level. For bitcoin, the coin type is 0. You can see the registered coin type for other coins here. <strong>Hardened derivation is used at this level.</strong></p><p><a href="https://github.com/satoshilabs/slips/blob/master/slip-0044.md">slips/slip-0044.md at master · satoshilabs/slips</a></p><h4>Account</h4><p>This level splits the key space into independent user identities, so the wallet never mixes the coins across different accounts. Accounts are numbered from index 0 in sequentially increasing manner. This number is used as child index in BIP32 derivation. <strong>Hardened derivation is used at this level.</strong></p><h4>Change</h4><p>Constant 0 is used for external chain and constant 1 for internal chain (also known as change addresses). External chain is used for addresses that are meant to be visible outside of the wallet (e.g. for receiving payments). Internal chain is used for addresses which are not meant to be visible outside of the wallet and is used for return transaction change. <strong>Public derivation is used at this level.</strong></p><h4>Index</h4><p>Addresses are numbered from index 0 in sequentially increasing manner. This number is used as child index in BIP32 derivation. <strong>Public derivation is used at this level.</strong></p><blockquote>We will be implementing BIP-32 algorithm from scratch in another article. For now, we will be using bip32 python package that’s available in pypi.</blockquote><p><a href="https://pypi.org/project/bip32/">bip32</a></p><p>In the previous section, we derived the account extended private and public root keys. Now we have to generate bip32 extended keys.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/656/1*SYr32NNBKx1JdenH0YM-Gw.png" /><figcaption>Generating BIP-32 extended private and public keys.</figcaption></figure><p>We supply the account private root key to the function, <em>from_xpriv()</em> from BIP32 class. If you check out the code for this function in the package, you could see that this function extracts information like network, depth, fingerprint, chaincode etc. from the extended root key to create a BIP32 wallet.</p><p>We then generate the BIP-32 extended public and private keys by supplying the BIP-32 path, <strong>m/44h/0h/0h/0 </strong>where <em>h</em> stands for hardened key.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/650/1*JcZ_0T3uHVhhvaRtF2CgvQ.png" /><figcaption>Generating BIP-44 Bitcoin Address.</figcaption></figure><p>To generate the address, we need to derive the private key from the BIP-32 extended private key and then public key from the private key.</p><p>To convert this public key to a bitcoin address, First we need to Perform sha-256 hashing on the public key, then Perform ripemd-160 hashing on the result of <strong>SHA-256</strong>. The version bytes are added in front of <strong>RIPEMD-160 hash</strong> (0x00 for Main Network).</p><p>We then used base58encode to encode the ripemd160 result into bitcoin address. Check more about base58 encoding here.</p><p><a href="https://en.bitcoin.it/wiki/Base58Check_encoding">Base58Check encoding</a></p><p>In short, base58 encoding performs another sha256 on the previous result two times and takes first 4 bytes of the second SHA-256 hash and add it at the end of extended RIPEMD-160 result that we previously calculated.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H8SkMaUjDxjc3F9jtURVtw.png" /><figcaption>Bitcoin address and extended keys.</figcaption></figure><p>Now, we can check if our output is correct.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*svur7jXpedptjlrG-WYUYw.png" /><figcaption>BIP32 Root Key</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_x4IWdMszeuhkWVcStXWzA.png" /><figcaption>BIP32 Extended Keys</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Z__Rv0laMMQrpIJZqGLjhg.png" /><figcaption>Bitcoin Address</figcaption></figure><p>As we can see that the Address and extended keys that we generated are correct and compatible.</p><h3>Conclusion</h3><p>Note that our bitcoin address starts with a <strong>‘1’</strong>. This means that our bitcoin address is a <strong>legacy address.</strong> Back when Bitcoin launched in 2009, this was the only way to create an address. Now there are other address formats like <strong>segwit</strong> and <strong>native segwit. </strong>We will see how to code those addresses in the next articles.</p><h3>Reference</h3><p><a href="https://github.com/bitcoin/bips">GitHub - bitcoin/bips: Bitcoin Improvement Proposals</a></p><blockquote>Join Coinmonks<a href="https://t.me/coincodecap"> Telegram Channel</a> and<a href="https://www.youtube.com/c/coinmonks/videos"> Youtube Channel</a> learn about crypto trading and investing</blockquote><h3>Also, Read</h3><ul><li><a href="https://coincodecap.com/crypto-trading-terminals">5 Best Crypto Trading Terminals</a> | <a href="https://coincodecap.com/best-defi-apps">Best DeFi Apps</a></li><li><a href="https://coincodecap.com/coinbase-vs-wazirx">Coinbase vs WazirX</a> |<a href="https://coincodecap.com/bitrue-review"> Bitrue Review</a> |<a href="https://coincodecap.com/poloniex-vs-bittrex"> Poloniex vs Bittrex</a></li><li><a href="https://coincodecap.com/crypto-exchanges-in-germany">Best Crypto Exchanges in Germany</a> |<a href="https://coincodecap.com/arbitrum"> Arbitrum: A Layer 2 Solution</a></li><li><a href="https://medium.com/coinmonks/binance-trading-bots-d0d57bb62c4c">Binance Trading Bots</a> |<a href="https://medium.com/coinmonks/okex-review-6b369304110f"> OKEx Review</a> |<a href="https://coincodecap.com/atani-review"> Atani Review</a></li><li><a href="https://medium.com/coinmonks/best-crypto-signals-telegram-5785cdbc4b2b">Best Crypto Trading Signals Telegram</a> |<a href="https://medium.com/coinmonks/moonxbt-review-6e4ab26d037"> MoonXBT Review</a></li><li><a href="https://coincodecap.com/buy-shiba-bitbns">How to buy Shiba(SHIB) Coin on Bitbns?</a> |<a href="https://coincodecap.com/buy-floki-inu-token"> Buy Floki</a></li><li><a href="https://coincodecap.com/coinflex-review">CoinFLEX Review</a> |<a href="https://coincodecap.com/aex-exchange-review"> AEX Exchange Review</a> |<a href="https://coincodecap.com/upbit-review"> UPbit Review</a></li><li><a href="https://coincodecap.com/best-cryptocurrency-blogs">10 Best Cryptocurrency Blogs</a> |<a href="https://coincodecap.com/youhodler-review"> YouHodler Review</a></li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4672e5057bb" width="1" height="1" alt=""><hr><p><a href="https://medium.com/geekculture/generate-bitcoin-wallet-address-using-bitcoin-improvement-proposal-44-4672e5057bb">Generate Bitcoin Wallet Address using Bitcoin Improvement Proposal 44</a> was originally published in <a href="https://medium.com/geekculture">Geek Culture</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>