<?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 Savan Padaliya on Medium]]></title>
        <description><![CDATA[Stories by Savan Padaliya on Medium]]></description>
        <link>https://medium.com/@padaliyasavan?source=rss-636926975ad------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*KUmLJ7YM4Qq0sH90QQFCDQ.jpeg</url>
            <title>Stories by Savan Padaliya on Medium</title>
            <link>https://medium.com/@padaliyasavan?source=rss-636926975ad------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 03:39:29 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@padaliyasavan/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[TypeScript for AI Applications: Stop Trusting LLM Responses]]></title>
            <link>https://padaliyasavan.medium.com/typescript-for-ai-applications-stop-trusting-llm-responses-dc1b73c26c79?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/dc1b73c26c79</guid>
            <category><![CDATA[openai]]></category>
            <category><![CDATA[typescript]]></category>
            <category><![CDATA[vertex-ai]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Thu, 07 May 2026 05:24:09 GMT</pubDate>
            <atom:updated>2026-05-07T05:24:09.835Z</atom:updated>
            <content:encoded><![CDATA[<p>You tweaked a prompt. The model started returning { score: “8” } instead of`{ score: 8 }. Your downstream code did math on a string. The result was wrong and you had no error, no warning, and no idea — until a user reported it three days later.</p><p>This is the most common failure mode in production AI apps. TypeScript can’t prevent it. But the combination of TypeScript and Zod can.</p><figure><img alt="TypeScript for AI Applications: Stop Trusting LLM Responses" src="https://cdn-images-1.medium.com/max/1024/1*R5XQQ4eZ-YFUgBiL5-4lpA.png" /></figure><h3>Why TypeScript Alone Isn’t Enough</h3><p>The OpenAI and Vertex AI SDKs are fully typed at the response <em>envelope</em> level status codes, token counts, model name. But the actual content your model generated is always <em>string</em>. TypeScript has no way to verify that the string <em>{“score”:8}</em> matches your interface at runtime.</p><p>Most developers hit this wall and write <em>as any</em>. That’s a time bomb.</p><p>The right pattern is two layers:</p><ol><li><strong>TypeScript interfaces</strong> — describe the shape you expect (compile-time)</li><li><strong>Zod schemas</strong> — verify the shape that actually arrived (runtime)</li></ol><p>Both layers are required. Skip either one and you’re still flying blind.</p><h3>Layer 1: Type Your SDK Calls Explicitly</h3><p>The OpenAI SDK exports its types. Import and use them instead of letting TypeScript infer <em>any </em>:</p><pre>import type {<br>  ChatCompletion,<br>  ChatCompletionMessageParam,<br>} from &#39;openai/resources/chat/completions&#39;;</pre><pre>const response: ChatCompletion = await client.chat.completions.create(params);<br>const content = response.choices[0]?.message?.content;<br>if (!content) throw new Error(&#39;Empty response from OpenAI&#39;);</pre><p>The <em>?. </em>chains are required — not defensive noise. The SDK types <em>choices[0] </em>and <em>message.content </em>as potentially <em>undefined </em>because an LLM can return a <em>content_filter </em>finish reason with no content. With <em>strict: true </em>in your <em>tsconfig.json </em>, the compiler forces you to handle this instead of silently propagating <em>undefined </em>into your business logic. <em>strict: true </em>is the single most important config change for any AI application. It enables <em>strictNullChecks</em>, which catches the exact class of runtime failures that hit you at 2am.</p><h3>Layer 2: Validate Structured Output with Zod</h3><p>When you need the model to return JSON, TypeScript interfaces aren’t enough. They only exist at compile time. Zod validates the actual runtime shape:</p><pre>import { z } from &#39;zod&#39;;</pre><pre>const ProductReviewSchema = z.object({<br>  sentiment: z.enum([&#39;positive&#39;, &#39;negative&#39;, &#39;neutral&#39;]),<br>  score: z.number().min(1).max(10),<br>  keyPoints: z.array(z.string()).max(5),<br>  recommendsPurchase: z.boolean(),<br>});</pre><pre>type ProductReview = z.infer&lt;typeof ProductReviewSchema&gt;;</pre><pre>ProductReviewSchema.parse(parsed)</pre><p>throws a <em>ZodError </em>with a field-level message if the model returns the wrong shape — string instead of number, missing field, out-of-range value. You catch it at the boundary.</p><pre>z.infer&lt;typeof ProductReviewSchema&gt;</pre><p>derives the TypeScript type from the schema. One source of truth. Change the schema and the type updates automatically — no duplication, no drift.</p><h3>Typed Prompt Builders</h3><p>Building prompts by string concatenation is refactor-proof in the worst way. A typed builder makes prompt construction testable and compiler-guided:</p><pre>interface SummariseRequest {<br>  text: string;<br>  maxWords: number;<br>  tone: &#39;formal&#39; | &#39;casual&#39; | &#39;bullet-points&#39;;<br>}</pre><pre>function buildSummarisePrompt(req: SummariseRequest): ChatCompletionMessageParam[] {<br>  return [<br>    { role: &#39;system&#39;, content: `Summarise in ${req.tone} tone, under ${req.maxWords} words.` },<br>    { role: &#39;user&#39;, content: req.text },<br>  ];<br>}</pre><p>Pass <em>‘informal’ </em>for <em>tone </em>— compiler error. Change the tone options single-point refactor. The prompt contract is now typed, not implicit.</p><h3>OpenAI JSON Mode + Zod Together</h3><p>For structured output, combine OpenAI’s native JSON mode with Zod into a reusable generic:</p><pre>async function structuredComplete&lt;T&gt;(<br>  messages: ChatCompletionMessageParam[],<br>  schema: z.ZodSchema&lt;T&gt;<br>): Promise&lt;T&gt; {<br>  const response = await client.chat.completions.create({<br>    model: &#39;gpt-4o&#39;,<br>    messages,<br>    response_format: { type: &#39;json_object&#39; },<br>    temperature: 0,<br>  });</pre><pre>const raw = response.choices[0]?.message?.content;<br>  if (!raw) throw new Error(&#39;Empty response&#39;);<br>  return schema.parse(JSON.parse(raw));<br>}</pre><p><em>temperature: 0</em> is correct for structured output — you want deterministic JSON, not creative variation in field names. The function is fully generic: the return type is inferred from the Zod schema passed in.</p><h3>The Failure You’re Avoiding</h3><p>Without this pattern, LLM output is an unverified <em>any </em>that flows through your entire codebase. The model changes its output shape — because of a prompt change, a model update, or context length pressure — and you find out from a user report, not a compiler error.</p><p>With TypeScript interfaces + Zod validation, you catch shape mismatches at the API boundary on every call. Runtime crashes become parse errors with clear field-level messages. The model’s unpredictability is contained.</p><p>I covered this in much more depth on my portfolio — typed streaming with <em>Stream&lt;ChatCompletionChunk&gt;</em>, Vertex AI typing patterns, a typed logger for tracking latency and token usage per call, and a guide on what to type carefully vs. where to keep it lightweight: <a href="https://savanpadaliya.com/blogs/typescript-for-ai-applications">Full article here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dc1b73c26c79" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Actually Happens When You Type a URL and Hit Enter?]]></title>
            <link>https://padaliyasavan.medium.com/what-actually-happens-when-you-type-a-url-and-hit-enter-d0bf18a5ad0b?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/d0bf18a5ad0b</guid>
            <category><![CDATA[backend-development]]></category>
            <category><![CDATA[system-design-concepts]]></category>
            <category><![CDATA[system-design-interview]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Fri, 24 Apr 2026 04:46:47 GMT</pubDate>
            <atom:updated>2026-04-24T04:46:47.518Z</atom:updated>
            <content:encoded><![CDATA[<p>Most developers answer this question with three steps: DNS resolves the domain, server sends back HTML, browser renders it. That’s the answer that passes a first-year CS quiz. It’s also the answer that ends your senior interview.</p><p>Here’s the version that actually matters.</p><h3><strong>The 3-Step Answer Leaves Out 12 Systems</strong></h3><p>When you type `github.com` and press Enter, your browser quietly orchestrates a chain of events before a single pixel renders on screen:</p><p>- <strong>DNS resolution</strong> across up to four server layers (your browser cache, OS cache, recursive resolver, authoritative nameserver)</p><p>- <strong>TCP 3-way handshake</strong> — a round-trip just to say “hello, I’m ready”</p><p>- <strong>TLS 1.3 handshake</strong>— negotiating encryption without ever sending the key over the wire</p><p>- <strong>HTTP request routing</strong> through a load balancer, past an nginx reverse proxy, into your app server</p><p>- <strong>Database query</strong>— connection pool checkout, query planning, buffer pool reads</p><p>- <strong>Response assembly</strong> — compression, cache headers, serialization</p><p>- <strong>Rendering</strong> — server-side, client-side, or static, each with different performance characteristics</p><p>All of this completes in under 300 milliseconds on a well-built site. When it doesn’t, engineers who understand this chain can pinpoint exactly which layer broke.</p><h3><strong>DNS: The Part Everyone Oversimplifies</strong></h3><p>DNS resolution isn’t one lookup. It’s a hierarchy:</p><p>Browser cache → OS cache → Router cache → ISP Recursive Resolver<br> → Root Nameserver → TLD Nameserver (.com) → Authoritative Nameserver → IP</p><p>Each layer caches the result based on a TTL. Your browser, your OS, your router — all of them might already have the answer. When they don’t, the recursive resolver walks the full hierarchy and caches the result for next time.</p><p>Why this matters in practice: Low TTL (60 seconds) means DNS changes propagate fast, but you’re hitting the resolver more often under load. High TTL (24 hours) reduces DNS traffic but means a bad record persists for a day. Production teams tune this based on how often they change infrastructure.</p><h3><strong>TCP + TLS: Two Handshakes Before Any Data Moves</strong></h3><p>TCP requires one round trip just to open a connection (SYN → SYN-ACK → ACK). On a 50ms connection, that’s 50ms of pure overhead. TLS 1.3 adds another round trip to negotiate encryption.</p><p>That’s why HTTP/2 multiplexes multiple requests over one connection and HTTP/3 (QUIC) eliminates the TCP handshake for repeat visitors. These aren’t academic optimizations — they’re the difference between a 200ms and 400ms TTFB.</p><h3><strong>Load Balancer → Web Server → App Server</strong></h3><p>Your request hits a load balancer first, which distributes traffic across a pool of servers using algorithms like round robin or least connections. It also health-checks backends and removes dead ones — this is how zero-downtime deploys work.</p><p>Then nginx (or Caddy, or Apache) handles the boring-but-important work: SSL termination, static file serving, compression, rate limiting. Dynamic requests get proxied to your app server, which runs your actual business logic.</p><p>The distinction matters because nginx can serve a static file in microseconds without touching your app. Your app server handling that same request costs 100x more.</p><h3><strong>The Database Is Usually the Problem</strong></h3><p>Your app server executes a query. What actually happens:</p><p>1. Checks out a connection from the pool (if no connections are free, the request waits — this is connection pool exhaustion)<br>2. The database plans the query — index scan vs sequential scan, join order, statistics<br>3. Executes against the buffer pool (in-memory first, disk if the page isn’t cached)<br>4. Serializes and returns the result</p><p>The N+1 problem lives here. ORMs generate one query to fetch a list, then one per item for related data. 100 items = 101 queries. Check your ORM’s query log in development before this becomes a production incident.</p><h3><strong>SSR vs CSR vs SSG</strong></h3><p>After the response arrives, rendering strategy determines what the user actually sees and when:</p><p>SSR (Server-Side Rendering): Server sends complete HTML. Fast first content paint. JavaScript hydrates it for interactivity.<br> CSR (Client-Side Rendering): Server sends a shell and a JS bundle. Browser builds the DOM. Slower first content paint, faster subsequent interactions.<br> SSG (Static Generation): HTML built at deploy time, served from CDN. Fastest possible TTFB, but only works for non-personalized content.</p><p>Modern frameworks give you per-page control. A marketing landing page can be SSG; a user dashboard needs SSR or CSR.</p><h3><strong>Why Senior Engineers Know This</strong></h3><p>When production is slow, junior engineers look at the code they wrote. Senior engineers look at the system. DNS latency, TCP RTT, database query plans, cache hit rates, render strategy — each of these is a lever. Knowing where each one lives means you can find the problem instead of guessing at it.</p><p>I covered the full version of this — with deeper dives into TLS handshakes, load balancer algorithms, database internals, and HTTP caching headers — on my portfolio. Worth a read if you want the whole picture: <a href="https://savanpadaliya.com/blogs/what-happens-when-you-type-a-url">Full article here</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d0bf18a5ad0b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Job Market Isn’t Killing Junior Developer Careers — This Mindset Is]]></title>
            <link>https://padaliyasavan.medium.com/the-job-market-isnt-killing-junior-developer-careers-this-mindset-is-34b7fb3f26a3?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/34b7fb3f26a3</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[junior-developer]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Mon, 20 Apr 2026 05:22:55 GMT</pubDate>
            <atom:updated>2026-04-20T05:22:55.700Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Originally published on </em><a href="https://savanpadaliya.com/blogs/junior-developer-job-market-fear"><em>Savan Padaliya’s portfolio</em></a></p><p>An undergrad student booked a session with me recently, genuinely scared. The question was: <em>”Companies hire 10 interns and release 8. What’s the point of even trying?”</em></p><p>I’ve heard versions of this question a dozen times in the past year. And every time, I notice the same thing — the fear is misdirected. The market isn’t the problem. The mindset is.</p><h3><strong>You’re Competing Against Clones</strong></h3><p>Here’s something brutal: most junior developers are building the exact same 50 projects. Todo app. Weather app. Netflix clone. Blog with CRUD.</p><p>They followed the same YouTube tutorials. Copied the same code. Pushed the same repos to GitHub.</p><p>Then they wonder why nobody’s impressed.</p><p>The recruiter has seen that repo 200 times this week. The GitHub star count doesn’t make it memorable. The fact that it’s identical to a hundred others does.</p><p>Tutorial projects teach you syntax. They don’t teach you how to think as an engineer.</p><h3><strong>The Driving Analogy That Changes Everything</strong></h3><p>Learning to drive on a quiet local street is comfortable. You know every turn, every stop sign. But put that same driver on a busy highway and they freeze.</p><p>That’s what tutorial-based learning produces — comfort within a known boundary. The moment the problem changes shape, the skills don’t transfer.</p><p>The only fix is to go beyond the block. Deliberately. And that means building things that aren’t on YouTube.</p><h3><strong>Build for Yourself, Not for a Resume</strong></h3><p>The most underrated advice I give junior developers: <strong>build something you actually need.</strong></p><p>Not a Netflix clone. Not another e-commerce frontend. Something that solves a real friction point in your daily life.</p><p>- A grocery list that remembers your recurring items</p><p>- An expense tracker that categorizes money the way <em>you</em> think about it</p><p>- A script that auto-organizes your downloads folder</p><p>- A price monitor that notifies you when something drops</p><p>Small scope. Real constraints. Genuine motivation.</p><p>That combination will teach you more than 10 tutorial projects because it forces you to make decisions — and decisions are what engineers are actually paid for.</p><p>When you can explain <em>why</em> you chose a particular data structure, <em>why</em> you redesigned the UI halfway through, <em>why</em> you handled that edge case differently — that’s what makes an interview memorable. You’re not showing a project. You’re showing how you think.</p><h3><strong>Ship It. Even If It Only Works on Localhost.</strong></h3><p>This is where most people lose.</p><p>You build something decent. It works on your machine. Then you think: <em>it’s not polished enough to post.</em> So it sits in a private repo forever and helps exactly nobody — including you.</p><p>Peter Thiel said it best in <em>Zero to One</em>: <strong>”If you have a product but don’t know how to sell it, you don’t have a product.”</strong></p><p>Most developers spend 90% of their time building and 10% on visibility. But in a market where every candidate has a GitHub, visibility is the differentiator.</p><p>A localhost app with a screen recording, two paragraphs explaining what you built and what you learned, posted on LinkedIn — that is more powerful than a polished app nobody knows exists.</p><p>You’re not marketing a product. You’re marketing your thinking.</p><h3><strong>The Framework That Actually Works</strong></h3><p>Here’s what I’ve seen consistently work for junior developers who break through:</p><p><strong>Build</strong> — something real that you actually use, something that forced decisions</p><p><strong>Document</strong> — write about why you built it, what you struggled with, what you’d change</p><p><strong>Ship</strong> — make it public, post about it, link it everywhere</p><p><strong>Engage</strong> — show up in communities, help others, get known</p><p><strong>Repeat</strong> — not once, over months</p><p>None of these steps are heroic. All of them compound.</p><p>The students who crack the 2-out-of-10 intern retention rate are rarely the smartest in the cohort. They’re the most visible — their thinking is on display, their work is documented, and their reputation is being built consistently.</p><h3><strong>The Market Has Room for This</strong></h3><p>The job market is tighter than 2021. That’s real. But “tight” doesn’t mean “closed” — it means the filter is higher.</p><p>And a real filter is good news if you’re willing to clear it. Because your competition is largely people who will follow tutorials, build clone repos, and wait to be discovered.</p><p>Build skills. Build real things. Tell people about them. Help others. Go beyond the block.</p><p>That’s not inspirational advice. That’s the actual playbook.</p><p><em>Savan Padaliya is a tech content creator covering AI, system design, and backend engineering. Follow for weekly insights on building scalable systems and developer career strategy.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=34b7fb3f26a3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Developers Are Experts at HOW. Nobody Told Us About WHY and WHAT.]]></title>
            <link>https://padaliyasavan.medium.com/developers-are-experts-at-how-nobody-told-us-about-why-and-what-30725801c445?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/30725801c445</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[what]]></category>
            <category><![CDATA[design-sprint]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Fri, 17 Apr 2026 05:08:11 GMT</pubDate>
            <atom:updated>2026-04-17T05:08:11.573Z</atom:updated>
            <content:encoded><![CDATA[<p>I recently participated in a Design Sprint — the 5-day product validation framework from Google Ventures.</p><p>My role going in: the engineer in the room.</p><p>What I walked out with: a completely different understanding of where most developers are missing the plot.</p><h3><strong>The HOW Trap</strong></h3><p>Developers are trained to answer HOW questions.</p><p>- HOW do we build this endpoint?</p><p>- HOW do we scale this database?</p><p>- HOW do we reduce latency?</p><p>These are great questions. But they’re downstream questions.</p><p>The harder ones — and the ones that actually determine whether the product succeeds — are:</p><p><strong>WHAT are we building?</strong></p><p><strong>WHY does it need to exist?</strong></p><p>Most of the time, those answers are handed to us as tickets. We never sit in the room where those decisions get made.</p><h3><strong>What a Design Sprint Actually Does</strong></h3><p>A Design Sprint is a structured 5-day process where cross-functional teams — product, design, engineering, business — align on a single problem, prototype one solution, and validate it with real users.</p><p>The entire first day is just understanding the problem space. No solutions. No code. Just mapping: who’s the user, what are they trying to do, where do things break down, what does success actually look like?</p><p>It’s the most important work that happens before any sprint planning board ever gets populated — and most engineers never see it.</p><h3><strong>What I Realized in That Room</strong></h3><p>Business and product teams are solving a fundamentally different problem than engineers.</p><p>They’re not optimizing systems. They’re identifying which problem is worth solving at all.</p><p>When you watch a product team debate WHAT to build, you start to understand why so much engineering effort gets thrown away. It’s not because the engineering was bad. It’s because the WHAT was wrong from the start.</p><p>And a wrong WHAT always produces a wrong HOW — no matter how clean the code is.</p><h3><strong>Why This Matters Right Now</strong></h3><p>With AI writing code faster every year, the HOW is becoming a commodity.</p><p>The engineers who will matter most aren’t the ones who can implement the fastest. They’re the ones who can ask the right questions before implementation starts.</p><p>Understanding WHY and WHAT isn’t a product skill. It’s the upstream half of engineering that most developers never develop.</p><p><em>*This is a condensed take. Read the full breakdown — including the Design Sprint structure, the WHAT problem, and three practical shifts I made after the sprint — on </em><a href="https://savanpadaliya.com/blogs/developers-missing-why-and-what"><em>my portfolio blog</em></a><em>.*</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=30725801c445" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Vertex AI vs AutoML: Choosing the Right AI Tool for Your Business]]></title>
            <link>https://padaliyasavan.medium.com/vertex-ai-vs-automl-choosing-the-right-ai-tool-for-your-business-113548c532f2?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/113548c532f2</guid>
            <category><![CDATA[vertex-ai]]></category>
            <category><![CDATA[automl]]></category>
            <category><![CDATA[google-cloud]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Tue, 08 Apr 2025 05:59:06 GMT</pubDate>
            <atom:updated>2025-04-08T05:59:06.142Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introduction</h3><p>AI is rapidly transforming SaaS businesses and startups, but choosing the right AI tool is crucial for success.</p><h4>Google Cloud offers two major AI solutions:</h4><p><strong>AutoML</strong> — A no-code AI tool for easy model training</p><p><strong>Vertex AI</strong> — A full-fledged AI platform for custom model development</p><h4>Which one is best for your business?</h4><p>In this guide, we’ll break down their key differences, features, and real-world use cases so you can make an informed decision.</p><h3>What is AutoML?</h3><p>AutoML (Automated Machine Learning) is designed for non-technical users and startups that want to build AI models without deep coding expertise.</p><h4>Why Choose AutoML?</h4><p><strong>No-Code AI</strong> — Train AI models with a simple drag-and-drop interface</p><p><strong>Fast Model Training</strong> — Ideal for startups that need quick AI-powered solutions</p><p><strong>Pre-Trained AI Models</strong> — Google provides AI models that require minimal customization</p><h4>Common Use Cases</h4><ul><li><strong>E-commerce</strong> — AI-powered product recommendations</li><li><strong>Customer Support </strong>— Sentiment analysis to classify customer queries</li><li><strong>Content Translation</strong> — AutoML Translation for multi-language SaaS applications</li></ul><p><strong>Example</strong>: A startup uses AutoML Vision to build an image recognition tool for detecting product defects in a manufacturing plant.</p><h3>What is Vertex AI?</h3><p>Vertex AI is Google Cloud’s end-to-end AI platform for companies that need full control over AI model training, deployment, and optimization.</p><h4>Why Choose Vertex AI?</h4><ul><li><strong>Custom Model Training </strong>— Build models from scratch using TensorFlow, PyTorch, or JAX</li><li><strong>Advanced AI Workflows </strong>— Train, deploy, and manage AI models in a scalable environment</li><li><strong>MLOps &amp; Automation</strong> — Automate data pipelines, model monitoring, and versioning</li></ul><h4>Common Use Cases</h4><ul><li><strong>Large-Scale AI Models</strong> — Training deep learning models for speech recognition</li><li><strong>Personalized AI Recommendations</strong> — AI-powered content recommendations for streaming platforms</li><li><strong>AI-Powered Forecasting</strong> — Predictive analytics for finance &amp; supply chain</li></ul><p><strong>Example:</strong> A SaaS company uses Vertex AI to build a custom fraud detection model for financial transactions.</p><h3>AutoML vs. Vertex AI: Feature Comparison</h3><p>Here’s a side-by-side comparison of AutoML and Vertex AI:</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/dd1f8a399b1217e312e99709967be7ba/href">https://medium.com/media/dd1f8a399b1217e312e99709967be7ba/href</a></iframe><h3>How to Choose the Right AI Tool for Your Business?</h3><h4>Use AutoML if:</h4><ul><li>You’re a SaaS startup or non-tech founder</li><li>You need AI models quickly without deep ML expertise</li><li>You want a cost-effective solution with minimal coding</li></ul><p><strong>Example</strong>: A customer support SaaS uses AutoML NLP to automate sentiment analysis and classify support tickets.</p><h4>Use Vertex AI if:</h4><ul><li>You have a technical team and need full AI model control</li><li>You need scalable, custom AI with deep learning capabilities</li><li>You require MLOps &amp; AI automation</li></ul><p><strong>Example</strong>: A streaming platform uses Vertex AI to build an AI-powered recommendation system for users.</p><h3>Conclusion &amp; Next Steps</h3><p>Choosing between AutoML and Vertex AI depends on your business needs, technical expertise, and scalability goals.</p><h4>What’s Next?</h4><p>Explore Google Cloud AI — Official Google AI documentation</p><p>Start experimenting with AI today!</p><p>Whether you’re a startup or an enterprise, Google Cloud has the right AI tool for you.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=113548c532f2" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Getting Started with AI on Google Cloud: A Beginner’s Guide]]></title>
            <link>https://padaliyasavan.medium.com/getting-started-with-ai-on-google-cloud-a-beginners-guide-eb8bb941ae8c?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/eb8bb941ae8c</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[automl]]></category>
            <category><![CDATA[googlecloudai]]></category>
            <category><![CDATA[vertex-ai]]></category>
            <category><![CDATA[google-cloud]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Wed, 26 Mar 2025 06:56:19 GMT</pubDate>
            <atom:updated>2025-03-26T06:56:19.535Z</atom:updated>
            <content:encoded><![CDATA[<h3>1. Introduction</h3><p>Artificial Intelligence (AI) is transforming the way businesses operate, especially in SaaS and startup ecosystems. From automating customer interactions to generating insights from big data, AI is a game-changer. However, setting up AI models can be complex without the right tools.</p><p>That’s where <strong>Google Cloud AI</strong> comes in. With <strong>scalable AI services, powerful APIs, and tools like Vertex AI and AutoML</strong>, Google Cloud makes AI accessible to everyone, from developers to non-technical founders.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xBhNnKH2tgPHZmI8D8dASw.jpeg" /></figure><h3>Why Choose Google Cloud for AI?</h3><ul><li><strong>Scalability:</strong> Run AI models efficiently with cloud infrastructure.</li><li><strong>Ease of Use:</strong> No-code/low-code solutions for faster deployment.</li><li><strong>Cost-Effective:</strong> Pay-as-you-go pricing with free-tier options.</li></ul><p>In this guide, we’ll walk through setting up AI on <strong>Google Cloud</strong>, from creating an account to deploying your first AI model. By the end, you’ll have a working AI solution running in the cloud.</p><h3>2. Google Cloud AI Basics</h3><p>Google Cloud offers multiple AI services, each catering to different use cases:</p><h3>Key AI Services on Google Cloud</h3><ol><li><strong>Vertex AI</strong> — A unified platform for training, deploying, and scaling machine learning models.</li><li><strong>AutoML</strong> — A no-code/low-code platform for training AI models without deep technical expertise.</li><li><strong>AI APIs</strong> — Ready-to-use APIs for:</li></ol><ul><li><strong>Vision API:</strong> Image recognition</li><li><strong>Speech-to-Text API:</strong> Audio transcription</li><li><strong>Natural Language API:</strong> Text analysis</li><li><strong>Translation API:</strong> Language translatio</li></ul><h3>How Does Google Cloud AI Compare to AWS &amp; Azure?</h3><ul><li><strong>Google Cloud AI</strong>: Best for ease of use, AI-powered APIs, and end-to-end model training.</li><li><strong>AWS AI</strong>: Offers a wide range of AI services but requires more configuration.</li><li><strong>Azure AI</strong>: Strong enterprise integration but has a steeper learning curve.</li></ul><h3>3. Setting Up AI on Google Cloud</h3><p>Before running an AI model, you need to <strong>set up Google Cloud AI services</strong>.</p><h3>Step 1: Create a Google Cloud Account</h3><ol><li>Visit <a href="https://console.cloud.google.com/welcome/new">Google Cloud Console</a>.</li><li>Sign up for a free-tier account (includes $300 in free credits).</li></ol><h3>Step 2: Enable AI Services &amp; APIs</h3><ol><li>Navigate to <strong>API &amp; Services</strong> in the Cloud Console.</li><li>Enable:</li></ol><ul><li>Vertex AI API</li><li>AutoML API</li><li>Cloud AI APIs (Vision, Speech, NLP, Translation)</li></ul><h3>Step 3: Set Up a Development Environment</h3><p>Choose from:</p><ul><li><strong>Cloud Shell</strong>: A web-based terminal in Google Cloud.</li><li><strong>Jupyter Notebook on AI Platform</strong>: Ideal for data scientists.</li><li><strong>Local Setup</strong>: Install the Google Cloud SDK on your machine.</li></ul><h3>4. Hands-on: Running Your First AI Model</h3><p>Let’s train an AI model using <strong>Google Cloud AutoML</strong>.</p><h3>Step 1: Prepare Your Dataset</h3><ol><li>Go to <strong>Vertex AI &gt; AutoML</strong> in Cloud Console.</li><li>Upload a dataset (e.g., images for classification).</li></ol><h3>Step 2: Train Your Model</h3><ol><li>Click <strong>New Model &gt; Train Model</strong>.</li><li>Select <strong>AutoML</strong> to let Google Cloud handle the training.</li></ol><h3>Step 3: Deploy &amp; Test</h3><ol><li>Deploy the trained model.</li><li>Use an API endpoint to send test data and receive predictions.</li></ol><p>Example API call to get a prediction:</p><pre>import requests</pre><pre>url = &quot;https://your-model-endpoint&quot;<br>data = {&quot;instances&quot;: [{&quot;image&quot;: &quot;your-image-url&quot;}]}<br>response = requests.post(url, json=data)<br>print(response.json())</pre><h3>5. Real-World Use Cases</h3><p>AI on Google Cloud is powering businesses across industries:</p><h3>Startups &amp; SaaS Companies Using Google Cloud AI</h3><ul><li><strong>E-commerce</strong>: Personalized recommendations with AI.</li><li><strong>Healthcare</strong>: AI-powered diagnostics using Vision API.</li><li><strong>Finance</strong>: Fraud detection using ML models.</li></ul><h3>Case Study: AI-Powered Customer Support</h3><p>A startup built an AI chatbot using <strong>Dialogflow</strong> (Google’s NLP API) to automate customer inquiries, reducing response times by 60%.</p><h3>6. Next Steps &amp; Resources</h3><p>Now that you’ve trained your first AI model, what’s next?</p><h3>Where to Go from Here?</h3><ul><li>Explore <strong>Google Cloud’s AI documentation</strong>: Google Cloud AI Docs</li><li>Learn <strong>how to optimize AI workloads</strong> for production in our Supporting Blog</li><li>Start experimenting with <strong>real-world AI use cases</strong>!</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=eb8bb941ae8c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is DeFi? or What is decentralized finance?]]></title>
            <link>https://padaliyasavan.medium.com/what-is-defi-or-what-is-decentralized-finance-3d8b9038d1bc?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d8b9038d1bc</guid>
            <category><![CDATA[decentralized-finance]]></category>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[defi]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Sat, 29 Oct 2022 06:23:55 GMT</pubDate>
            <atom:updated>2022-10-29T06:23:55.021Z</atom:updated>
            <content:encoded><![CDATA[<p>In this article, we are going to find out what is DeFi? or what is decentralized finance. Before that, let’s find out why we need decentralized finance &amp; what is the problem with centralized finance or the current world financial system.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0ZtcNyddnszak9QIoBuooQ.png" /><figcaption>Ref: <a href="https://tinyurl.com/vsbxxaa2">https://tinyurl.com/vsbxxaa2</a></figcaption></figure><h3><strong>Why do we need decentralized finance(DeFi)?</strong></h3><p>In centralized finance, central authorities like banks and governments rule over our money. Let’s take a scenario when the government rolls out a rule to the user that the user can’t withdraw money from their respective banks. What will happen in this scenario? User is unable to access their own money. In some scenarios, the user pays hefty fees to make transactions.</p><p>Users need to follow rules and regulations of the government and banks. This is the problem with financial organizations run by central authorities. Now let’s check how decentralized finance can solve these problems.</p><h3><strong>What is decentralized finance(DeFi)?</strong></h3><p>Decentralized finance uses blockchain technology which helps in eliminating central authorities, through a peer-to-peer connection that uses security protocols, connectivity software, and hardware advancements.</p><p>Decentralized finance(DeFi) gives freedom to users and more control over their money through personalized wallets and trading services.</p><h3><strong>How does decentralized finance(DeFi) work?</strong></h3><p>Decentralized finance applications use blockchain technology. Computer program handles these applications, which are responsible for recording user information and transactions on a blockchain and that is known as Smart Contract.</p><p>Smart Contracts are not smart at all. It is only a piece of code, that has several conditions to follow and performs operations basis on that conditions. A user interface is created, which helps the user to perform operations easily. That interface interacts with smart contracts.</p><h3><strong>Advantages and Disadvantages of decentralized finance(DeFi).</strong></h3><h4>Advantages</h4><ul><li>Decentralized applications help the user to invest, trade, and transfer capital all over the globe without intermediaries.</li><li>Organizations can raise funds with IEOs or ICOs or IDOs, by offering tokens as proof of investments like IPOs.</li></ul><h4>Disadvantages</h4><ul><li>Participation in decentralized finance is complex and very hard to understand.</li><li>Very high volatility in token pricing</li></ul><h3>Big Story Short</h3><p>Decentralized Finance is the future of the financial system, which will bring the solution to several hurdles like central authority ruling and a hefty amount of transaction fees which vary from bank to bank and country to country.</p><p>Decentralized finance uses blockchain protocol and promotes peer-to-peer transactions. Blockchain protocols help in creating a secure and transparent financial system. There is a risk of a smart contract hack whenever code is not written with proper securities.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d8b9038d1bc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is the consensus mechanism in blockchain?]]></title>
            <link>https://padaliyasavan.medium.com/what-is-the-consensus-mechanism-in-blockchain-c16523716738?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/c16523716738</guid>
            <category><![CDATA[blockchain]]></category>
            <category><![CDATA[consensus-algorithm]]></category>
            <category><![CDATA[consensus-mechanism]]></category>
            <category><![CDATA[proof-of-work]]></category>
            <category><![CDATA[proof-of-stake]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Mon, 03 Oct 2022 08:32:08 GMT</pubDate>
            <atom:updated>2022-10-03T08:32:08.307Z</atom:updated>
            <content:encoded><![CDATA[<p>Consensus means a general agreement in simple words. The consensus mechanism in the blockchain is a process. In which peers of the blockchain network reach an agreement about the present state of the data in a network.</p><p>Consensus mechanisms are used to verify transactions and maintain the security of the underlying blockchain.</p><p>There are two widely used consensus mechanisms in recent times.</p><p>1) Proof of Work (PoW)</p><p>2) Proof of Stake (PoS)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EMCaPZI3zPbG8HWzOyZrUw.png" /><figcaption>Source: <a href="https://tinyurl.com/fxtf5u5u">https://tinyurl.com/fxtf5u5u</a></figcaption></figure><h3><strong>Chainalysis</strong></h3><p>Roughly there are two categories of consensus mechanism exists:</p><p>1) Proof-based, leader based, or the Nakamoto consensus whereby a leader is elected and proposes a final value.</p><p>2) Byzantine fault tolerance-based, which is a more traditional approach based on rounds of votes.</p><h3><strong>Proof Of Work (PoW)</strong></h3><p>Proof of work is a form of cryptographic proof in which one party proves to others that a certain amount of a specific computational effort has been expected.</p><p>Verifiers can subsequently confirm this expenditure with minimal effort on their part.</p><p>Proof of work was later popularized by Bitcoin, as a foundation for consensus in a permissionless decentralized network, in which miners compete to append blocks and mint new currency, each miner experiencing a success probability proportional to the computational effort expended.</p><h3><strong>Proof of Stake (PoS)</strong></h3><p>With proof of stake, cryptocurrency owners validate block transactions based on the number of stacked coins.</p><p>Proof of work was the initial consensus mechanism used to validate a blockchain and add new blocks, whereas proof of stake was created as an alternate option to proof of work.</p><p>Miners need to solve the cryptographic puzzle in the proof of work mechanism, whereas the proof of stake mechanism requires to stake and hold cryptocurrency to earn incentives.</p><p>The next block writer on the ledger is randomly selected with odds in favor of nodes with the largest stack positions.</p><p><strong>Trivia:</strong> Ethereum protocol switched consensus mechanism from PoW to PoS on September 15th, 2022, because it is more secure, less energy-intensive, and better for implementing new scaling solutions, compared to the previous proof-of-work architecture.</p><p>The consensus mechanism used in bitcoin is Proof Of Work, and the consensus mechanism used in Ethereum is Proof Of Stack.</p><p><strong>Proof of History(PoH) </strong>is a whole different term, which is a Solana consensus mechanism. Solana is a public blockchain that is second on the list after Ethereum in decentralized application development.</p><p>The Proof of History provides a way to cryptographically prove the passage of time and where events fall in that timeline. This consensus mechanism is used in tandem with another more conventional algorithm like the Proof of Work (PoW) or Proof of Stake (PoS). The Proof of History makes the Proof of Stake more efficient and resilient in Solana.</p><p><em>References:</em></p><p><a href="https://en.wikipedia.org/wiki/Proof_of_work"><em>https://en.wikipedia.org/wiki/Proof_of_work</em></a></p><p><a href="https://www.investopedia.com/terms/p/proof-stake-pos.asp"><em>https://www.investopedia.com/terms/p/proof-stake-pos.asp</em></a></p><p><a href="https://www.infoworld.com/article/3666736/solana-blockchain-and-the-proof-of-history.html"><em>https://www.infoworld.com/article/3666736/solana-blockchain-and-the-proof-of-history.html</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c16523716738" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is a blockchain?]]></title>
            <link>https://medium.com/coinmonks/what-is-a-blockchain-fe0e42af9968?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/fe0e42af9968</guid>
            <category><![CDATA[blockchain-development]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Sun, 17 Jul 2022 14:16:30 GMT</pubDate>
            <atom:updated>2022-10-03T14:57:25.779Z</atom:updated>
            <content:encoded><![CDATA[<p>In this current digital age, blockchain is the future of technology. So what is blockchain? Let me share some information regarding the blockchain I have learned so far.</p><p>Blockchain has multiple definitions. Let’s go through the simplest one. <strong>Blockchain is a distributed shared ledger, which can be considered as a shared ledger of transactions, transactions are ordered and grouped in blocks</strong>. In current real-world model is based on a private database that each organization maintains. A distributed ledger, it serves as a single source of truth for all members that are using the blockchain. Blockchain is basically a linked list that uses hash pointers instead of normal pointers. Hash pointers are used to point to the previous block. The structure of blockchain can be visualized as shown in the following diagram.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/850/1*xJFs2K_b22wc6T603FEa2w.png" /></figure><h3>Elements of blockchain.</h3><p>Every block has an address, hash, transaction, nonce, timestamp, etc. There are other elements as well, but these 4 are major and we are going to discuss that in this article.</p><p><strong>Addresses </strong>are unique identifiers that are used in a transaction on the blockchain to denote sender and receivers. An address is usually a public key or derived from the public key.</p><p><strong>Hash </strong>is an encrypted hash of all data included in a block like the address of the previous block, nonce, transaction, timestamp, etc. In block we store has of the previous block to maintain the immutability of the blockchain.</p><p><strong>Nonce </strong>is a random number used to create a hash between the desired target.</p><p><strong>Timestamp</strong> is a timestamp of the block when it was mined.</p><h3>Feature of Blockchain.</h3><p>Blockchain provides <strong>security</strong>. As data of blockchain is available to nodes of blockchain and they are responsible for adding a block to the blockchain after approval by all nodes by using a consensus mechanism it is too much difficult to mess with data.</p><p><strong>Immutability </strong>is another key feature of blockchain: records once added to the blockchain are immutable. to reverse the changes of data on the blockchain is almost impossible as it requires an unaffordable amount of computational power.</p><p><strong>Uniqueness</strong>: blockchain ensures that every transaction ins unique and has not been spent already.</p><p>There are many more features of blockchain available. I have shared the only 3 features of blockchain right now.</p><p>References: Mastering Blockchain book.</p><blockquote>New to trading? Try <a href="https://medium.com/coinmonks/crypto-trading-bot-c2ffce8acb2a">crypto trading bots</a> or <a href="https://medium.com/coinmonks/top-10-crypto-copy-trading-platforms-for-beginners-d0c37c7d698c">copy trading</a></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fe0e42af9968" width="1" height="1" alt=""><hr><p><a href="https://medium.com/coinmonks/what-is-a-blockchain-fe0e42af9968">What is a blockchain?</a> was originally published in <a href="https://medium.com/coinmonks">Coinmonks</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What is decentralization?]]></title>
            <link>https://padaliyasavan.medium.com/what-is-decentralization-7650eed8d4c1?source=rss-636926975ad------2</link>
            <guid isPermaLink="false">https://medium.com/p/7650eed8d4c1</guid>
            <category><![CDATA[web3]]></category>
            <category><![CDATA[decentralization]]></category>
            <category><![CDATA[distributed]]></category>
            <category><![CDATA[blockchain]]></category>
            <dc:creator><![CDATA[Savan Padaliya]]></dc:creator>
            <pubDate>Sun, 03 Apr 2022 09:35:59 GMT</pubDate>
            <atom:updated>2022-04-03T09:35:59.571Z</atom:updated>
            <content:encoded><![CDATA[<p>Decentralization is one of the most used terms or a word in web3 space. So, what is decentralization? In simple words, it is just passing control or any entity to multiple entities rather than just one entity. It’s still complex right. Let’s break it down with an example. Let’s use a database as an example. Google, Meta(previously known as Facebook), etc. uses a database. Their database is managed at some central system, which means some specific storage device or server room which can be termed centralization. In the decentralized database, data is stored and managed by multiple servers(nodes), which are not controlled by any organization.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*sNx-Ee6tnKKf2rmoXFZmFw.png" /></figure><h3>How can be decentralized achieved?</h3><p>Let’s use the database as an example. Whenever anyone uploads any data to a decentralized database then they inform it to the system and the process of database entry is informed to all servers (nodes) that are responsible for database management. All servers(nodes) now work on database entry. Whenever an entry is done successfully then that entry is informed to all peer servers(nodes) that the entry has been completed. Now they start working on another entry. This is the process in simple words how decentralization is achieved. In web3 specifically, thousands of hours of research and billions of hash power are being used just to achieve decentralization.</p><h3>What is the difference between decentralized and distributed?</h3><p>Decentralization means that not one single entity has control over all the processing done in the system. Whereas distributed means not all processing of the transaction is done in the same place.</p><h3>Why decentralization is used?</h3><p>There are several reasons for using decentralization and a major reason decentralization is being used in web3 space.<br>1) Fault tolerance<br>2) Attack resistance<br>3) Collusion resistance</p><ol><li>Fault Tolerance<br>A decentralized system is not likely to fail, as it relies on many individual components that are not likely to fail accidentally. So there is very very likely to fail fault tolerance is very negligible.</li><li>Attack resistance<br>To attack decentralized systems and manipulate systems is highly expensive as they lack a sensitive central point to attack over and manipulate.</li><li>Collusion resistance<br>It is much harder for individuals in a decentralized system to collude for their benefit. It is managed in such a way that harms very less to a well-coordinated system.</li></ol><h3>Conclusion</h3><p>It is too much hard to achieve decentralization as it required a huge and well-maintained ecosystem to maintain it and make it fault resistance, attack resistance, and Collusion resistance. Once the system is developed then there are servers (nodes) to manage it and maintain decentralization on its own. It’s too much expensive to break down or take over a decentralized system.</p><h3>References</h3><p><a href="https://medium.com/@VitalikButerin/the-meaning-of-decentralization-a0c92b76a274">The Meaning of Decentralization</a></p><p>Image: <a href="https://steemit.com/steemit/@jholdsworthy/decentralization-is-the-future">https://steemit.com/steemit/@jholdsworthy/decentralization-is-the-future</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7650eed8d4c1" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>