<?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 Bharath Thakur on Medium]]></title>
        <description><![CDATA[Stories by Bharath Thakur on Medium]]></description>
        <link>https://medium.com/@bharaththakur_91308?source=rss-473be74f4400------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*pksT8cgjauZeFtN3.jpg</url>
            <title>Stories by Bharath Thakur on Medium</title>
            <link>https://medium.com/@bharaththakur_91308?source=rss-473be74f4400------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 11:42:33 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@bharaththakur_91308/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[Building a Multi-Sided P2P Lending Platform: Where LOS and LMS Stop Being Theory]]></title>
            <link>https://medium.com/@bharaththakur_91308/building-a-multi-sided-p2p-lending-platform-where-los-and-lms-stop-being-theory-33ece04f0c15?source=rss-473be74f4400------2</link>
            <guid isPermaLink="false">https://medium.com/p/33ece04f0c15</guid>
            <category><![CDATA[lending]]></category>
            <category><![CDATA[software-architecture]]></category>
            <category><![CDATA[p2p-lending]]></category>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[clean-architecture]]></category>
            <dc:creator><![CDATA[Bharath Thakur]]></dc:creator>
            <pubDate>Wed, 07 Jan 2026 07:31:34 GMT</pubDate>
            <atom:updated>2026-01-07T07:31:34.112Z</atom:updated>
            <content:encoded><![CDATA[<p>In the first article, I wrote about how systems that make irreversible decisions naturally split into two phases: <strong>Decision Time (LOS)</strong> and <strong>Consequence Time (LMS)</strong>.</p><p>I built a P2P lending platform at <strong>Arthmate</strong> that tested whether that separation actually holds once real constraints show up.</p><p>On paper, the split looks clean.<br>In reality, it only survives if the platform is forced to respect it.</p><p>At Arthmate, we scaled this to thousands of investors, dozen of lending partners, and hundreds of crores in capital deployed. At that scale, architecture stops being an abstract discussion. You can’t rely on manual fixes or institutional memory. Whatever boundaries you designed either hold, or they fail loudly.</p><p>P2P lending is one of the fastest ways I know to find out whether your platform thinking is real or just well-articulated.</p><p>It adds three constraints that remove all room for optimism:</p><ul><li>Regulatory limits that cannot be violated</li><li>Partner ecosystems with very different incentives</li><li>Multi-party ledgers where money has to reconcile exactly</li></ul><p>If you’ve worked on billing systems, ad attribution, marketplaces, or settlement platforms, the patterns will feel familiar. Lending just makes the consequences harder to ignore.</p><h3>The Three-Sided Problem</h3><p>Most platforms are two-sided.</p><p>Payment networks connect merchants and cardholders.<br>Cloud platforms connect developers and infrastructure.<br>Job boards connect candidates and employers.</p><p>P2P lending isn’t like that. It’s <strong>three-sided</strong>, and that changes the shape of the system in ways that aren’t obvious upfront.</p><p>On one side, you have <strong>investors</strong>.<br>They provide capital. They operate under RBI limits. They expect predictable returns. They might come directly through your app, or they might arrive via Independent Financial Advisors who expect commissions.</p><p>On another side, you have <strong>borrowers</strong>.<br>They need capital. They usually come through fintech partners who expect margins. They have to be underwritten, disbursed to, and collected from over long periods of time.</p><p>And in the middle sits <strong>the platform</strong>.</p><p>The platform isn’t just matching supply and demand. It is creating irreversible contracts between parties who will never interact directly, enforcing those contracts over time, and making sure that when money moves, it moves correctly.</p><p>If you get this wrong, you don’t just lose trust.<br>You lose the ability to operate at all.</p><h3>Supply Side: Investor Acquisition Without Losing Control</h3><p>Investors are easier to reason about than borrowers, but much harder to control once you scale.</p><p>Here’s why.</p><p>Investors don’t arrive through a single funnel. Some come directly via your app. Others come through Independent Financial Advisors.</p><p>The Reserve Bank of India sets hard limits.<br>₹10 lakh per investor. ₹50 lakh if the investor has a CA-verified net-worth certificate.</p><p>Those limits apply across all channels.</p><p>If the same person signs up through your app and through three different IFAs, they still have one limit. Not four.</p><p>That immediately forces a <strong>centralized investor registry</strong>.</p><p>Every onboarding flow, no matter where it originates, has to check that registry first. If the investor already exists, you reject the duplicate. If they’re new, you create a single identifier and track every rupee they deploy across the ecosystem.</p><p>This isn’t a UX choice. It’s table stakes for regulatory compliance.</p><p>The part many teams underestimate is the IFA layer.</p><p>IFAs aren’t employees. They’re independent businesses with their own economics. They bring investors to your platform and expect to be paid correctly.</p><p>Commissions are usually slab-based on Assets Under Management. The more AUM an IFA brings in, the higher their payout percentage. In many setups, payouts only trigger after certain thresholds are crossed.</p><p>That means the platform has to know, at any point:</p><ul><li>Which IFA sourced which investor</li><li>How much that investor has actually deployed</li><li>What returns they’ve earned</li><li>What commission is owed and when</li></ul><p>And because IFAs operate independently, they will reconcile your numbers against their own. If your ledger doesn’t line up with theirs, trust breaks immediately.</p><p>There’s one more constraint that matters here.</p><p><strong>Investors don’t choose borrowers.</strong></p><p>They choose a plan. Tenure, interest rate, payout structure. The platform pools capital and allocates it to borrowers based on risk, availability, and liquidity.</p><p>That’s not a limitation. It’s what makes P2P workable at scale.</p><p>If investors picked individual borrowers, liquidity would collapse and the system would turn into a fragile marketplace.</p><p>Pooling works, but it comes with responsibility.</p><p>At any point in time, the platform has to be able to explain where every rupee is, which borrowers it went to, and what returns it’s generating.</p><p>Investor onboarding owns the <strong>decision</strong>.<br>Can this person invest? How much capacity do they have left? Who sourced them?</p><p>Investor repayments own the <strong>consequence</strong>.<br>Tracking returns, splitting repayments across loans, and making sure the ledger always reconciles.</p><p>Once those two responsibilities blur, operations ends up fixing statements instead of running the platform.</p><h3>Demand Side: Partner Integration Without Accumulating Debt</h3><p>Borrowers don’t come to you directly at scale. They come through partners.</p><p>Fintech companies, NBFCs, lending apps. Businesses with distribution and borrower relationships, but not always their own capital.</p><p>Each partner has its own margin structure, often referred to as the <strong>huddle rate</strong>.</p><p>If a borrower is charged 18 percent and investors are promised 12 percent, the remaining 6 percent is split between the platform and the partner based on their agreement.</p><p>Partner A might keep 2 percent.<br>Partner B might keep 3 percent.<br>Partner C might accept less because they bring better borrowers.</p><p>None of this is standardized.</p><p>Where many platforms get into trouble is letting every partner build a custom integration.</p><p>Different schemas. Different assumptions. Different edge cases.</p><p>Six months in, engineering is maintaining a mess of special-case logic, and every new partner increases fragility.</p><p>The scalable approach isn’t clever. It’s disciplined.</p><p>The platform exposes <strong>standardized APIs</strong>.</p><p>Same onboarding flow. Same data schema. Same error handling. Partner-specific behavior is handled through configuration, not bespoke code.</p><p>This takes more work upfront. But it’s what allows you to grow from a handful of partners to dozens without drowning in tech debt.</p><p>The platform also owns <strong>underwriting</strong>.</p><p>Partners source borrowers and provide context. The decision to lend stays with the platform.</p><p>That’s not about control. It’s about accountability.</p><p>If a borrower defaults, investors don’t blame the partner. They blame the platform.</p><p>Centralizing underwriting also centralizes learning. When risk models change, every partner benefits immediately. Decentralized underwriting leads to fragmented risk and slow feedback loops. Neither scales.</p><p>Once a borrower is approved, the platform assigns a <strong>Loan ID</strong>.</p><p>Partners may track loans using temporary identifiers. Internally, every loan gets a unique, immutable ID that survives the entire lifecycle.</p><p>This matters because borrowers can take multiple loans, sometimes through different partners. Each obligation has to remain cleanly separable.</p><p>LOS handles onboarding, underwriting, partner integrations, and the decision to disburse.<br>LMS handles repayments, collections, defaults, and splitting money back to investors.</p><p>The boundary stays strict, even when it’s tempting to blur it.</p><h3>The Control Layer: Asset Liability Management Is Not Optional</h3><p>Most fintech products treat finance as a back-office concern.</p><p>In P2P lending, <strong>finance is the platform</strong>.</p><p>Asset Liability Management is where the platform decides how much capital to allocate, whether liquidity is sufficient, and whether margins are real or just assumed.</p><p>ALM has to model:</p><ul><li>Investor inflows</li><li>Partner allocations</li><li>Tech and vendor costs</li><li>Operational overhead</li><li>IFA commissions</li><li>Partner margins</li><li>Risk provisions</li></ul><p>If any of these are wrong, the impact shows up later, when it’s expensive to fix.</p><p>Building ALM wasn’t just a technical exercise. It required convincing finance and legal that this wasn’t optional overhead, but a core platform capability. Without that alignment, the system would have shipped with blind spots that only appear under audit or stress.</p><p>ALM is also where discipline gets enforced.</p><p>Partners who underperform get throttled. Partners who perform well get more allocation. Liquidity buffers get adjusted as conditions change.</p><p>This isn’t fully automated. Judgment is involved. But without clear visibility, the platform is flying blind.</p><p>You can have capital and still be unable to deploy it safely.<br>You can believe margins are healthy and discover too late that they weren’t.</p><h3>Operations and Compliance Are Load-Bearing</h3><p>Two layers that are easy to underestimate.</p><p><strong>Operations</strong> handles what automation can’t. Edge cases, anomalies, investigations.</p><p>An investor document doesn’t quite match. Ops reviews it.<br>An IFA shows suspicious behavior. Ops investigates.<br>A borrower appears twice via different partners. Ops resolves it.</p><p>Without ops, small issues accumulate quietly until they become systemic.</p><p><strong>Compliance</strong> ensures the platform can answer uncomfortable questions years later.</p><p>Why was this loan approved?<br>How was this investor limit calculated?<br>Where did this commission payment go?<br>Why wasn’t this default covered by insurance?</p><p>Compliance isn’t a checkbox. It’s an ongoing obligation backed by data captured at decision time.</p><h3>Why This Is a Platform Maturity Test</h3><p>P2P lending isn’t complex because of the business model.</p><p>It’s complex because it forces platform boundaries to be correct from day one.</p><p>LOS has to produce decisions that survive for years.<br>LMS has to enforce consequences without drift.<br>ALM has to surface reality, not optimism.<br>Operations and Compliance have to catch what everyone else misses.</p><p>Every decision is irreversible.<br>Every consequence lasts for the loan lifecycle.</p><p>There’s no “we’ll clean this up later.”</p><p>P2P lending is unforgiving. But if you build it well, it teaches you how to design platforms for any domain where irreversible state, multi-party coordination, and long-term consequences matter.</p><p>The same patterns show up in payment settlement systems, advertising attribution platforms, and marketplace guarantees. Anywhere a decision creates obligations that outlive a single request, these boundaries matter.</p><p>Get them right, and the platform scales without accumulating invisible risk.<br>Get them wrong, and you spend years unwinding decisions made when context was cheap and consequences felt distant.</p><h3>What Comes Next</h3><p>This article focused on architecture, boundaries, and responsibility.</p><p>The next one will go deeper into why LOS platforms break as they scale, how delayed feedback loops poison decision quality, and what most PMs miss when designing growth systems where mistakes can’t be undone.</p><p>But before you worry about growth, you need to understand the platform.</p><p>Now you do.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=33ece04f0c15" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Why Systems That Make Irreversible Decisions Break Differently]]></title>
            <link>https://medium.com/@bharaththakur_91308/why-systems-that-make-irreversible-decisions-break-differently-7b552b62d2d4?source=rss-473be74f4400------2</link>
            <guid isPermaLink="false">https://medium.com/p/7b552b62d2d4</guid>
            <category><![CDATA[saas]]></category>
            <category><![CDATA[fintech]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Bharath Thakur]]></dc:creator>
            <pubDate>Mon, 05 Jan 2026 13:11:54 GMT</pubDate>
            <atom:updated>2026-01-05T13:11:54.050Z</atom:updated>
            <content:encoded><![CDATA[<h3>Lessons from Lending at Scale</h3><p>At scale, some systems make decisions that cannot be undone.</p><p>Not soft failures you can retry.<br>Not experiments you can roll back.</p><p>Irreversible state.</p><p>Money disbursed.<br>Contracts signed.<br>Trades executed.</p><p>Once this happens, time stops being neutral. It becomes a source of risk.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*sJ-bYQgJg44f5aCZnYLfEQ.png" /></figure><p>This article is about how platforms behave after they cross that line, using lending as a concrete case study. The pattern applies far beyond fintech.</p><h3>The Core Constraint: Decision Time vs Consequence Time</h3><p>Most platforms are built for iteration.</p><p><strong>Ship fast. Measure. Adjust. Recover.</strong></p><p>Systems that create irreversible state cannot rely on that loop. They naturally split into two very different responsibilities:</p><ol><li>A system that makes a high-stakes decision in a narrow time window</li><li>A system that enforces the consequences of that decision over months or years</li></ol><p>In lending, these show up as:</p><ul><li><strong>Loan Origination Systems (LOS)</strong></li><li><strong>Loan Management Systems (LMS)</strong></li></ul><p>The names are domain-specific. The separation is universal.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/568/1*cK2oNdOFPjUCXCPNw7Mmqg.avif" /></figure><h3>Decision-Time Systems (LOS in Lending)</h3><p>Decision-time systems operate under compression.</p><p>They aggregate many signals, from many external sources, under tight latency constraints, to produce a single outcome that cannot be rewritten later.</p><p>In lending, that decision answers questions like:</p><ul><li>Should we proceed?</li><li>On what terms?</li><li>Under which policy version?</li><li>On whose behalf?</li></ul><p>The data itself is not the point. The <strong>decision contract</strong> is. Once that contract exists, every downstream system assumes it was correct. Good decision-time platforms obsess over:</p><ul><li>Explicit policy boundaries</li><li>Versioned logic</li><li>Explainable outcomes</li><li>Clear attribution and ownership</li></ul><p>Not because it feels elegant, but because six months later someone will ask why a decision was made. “The system said so” is never a sufficient answer.</p><p><strong>Transferable principle</strong><br>Any system that makes irreversible decisions needs durable decision contracts. If you cannot reconstruct the inputs, policies, and ownership behind an outcome, you are accumulating invisible risk.</p><h3>Why Decision-Time Systems Encourage Dangerous Optimism</h3><p>Decision-time systems are where everything looks good.</p><p>You add a new signal.<br>You relax a threshold.<br>You onboard a new partner.</p><p>Conversion improves within days. Dashboards turn green. The change ships quietly into production.</p><p>Nothing breaks.</p><p>Three months later, a different meeting happens.</p><p>Delinquencies tick up, but not enough to panic.<br>Collections asks for more tooling.<br>Finance flags a growing gap between expected and realized revenue.</p><p>Six months later, the questions get sharper.</p><p>Why are write-offs concentrated in a specific cohort?<br>Which policy version approved these accounts?<br>Was this partner always this risky, or did something change?</p><p>By this point, the original decision is cold. The PM has moved teams. The context lives in Slack threads no one can find. The system can tell you <em>what</em> happened, but not <em>why</em>.</p><p>This is the dangerous part.</p><p>Decision-time systems reward speed immediately and punish mistakes slowly, diffusely, and out of band. The people who feel the pain are rarely the ones who shipped the change.</p><p>That is why decision-time systems cannot be treated as funnels or flows. They must be treated as platforms, with the same rigor you would apply to billing, ads, or access control.</p><p>Delayed failures are still failures. They are just harder to trace, harder to fix, and far more expensive.</p><h3>Consequence-Time Systems (LMS in Lending)</h3><p>Once an irreversible decision is executed, the platform enters consequence time.</p><p>These systems do not care why a decision was made. They care that the world behaves consistently afterward.</p><p>In lending, that means:</p><ul><li>Time-based state transitions</li><li>Interest and fee accrual</li><li>Partial failures and retries</li><li>External reconciliation</li><li>Auditability years later</li></ul><p>This is why LMS platforms feel conservative and slow. They resemble billing systems more than consumer products.</p><p>They exist to preserve trust.</p><p>Most failures here are not outages. They are small inconsistencies that compound quietly until finance, legal, or regulators stop believing the system.</p><p>When that happens, roadmap freedom disappears.</p><p><strong>Transferable principle</strong><br>Any system enforcing irreversible decisions over time must prioritize correctness over flexibility. Once trust erodes, every change becomes a negotiation.</p><h3>Why Partners Change the Equation</h3><p>Once multiple parties depend on the same irreversible decision, ambiguity becomes toxic.</p><p>Modern platforms rarely act alone. They operate in ecosystems of partners with different incentives, failure modes, and tolerance for risk.</p><p>In lending, this includes sourcing partners, co-lenders, collection agencies, and investors. In other domains, it looks like marketplaces, advertisers, or third-party developers.</p><p>Platforms that treat partners as simple integrations struggle. Platforms that treat partners as first-class participants invest in explicit contracts, observable events, and clear ownership boundaries.</p><p>This is not an integration problem.<br>It is a platform design problem.</p><h3>Why P2P and Multi-Party Models Expose Weak Platforms</h3><p>P2P and multi-lender models remove the ability to hide behind averages.</p><p>One decision can affect multiple principals at once. That forces:</p><ul><li>Precise attribution</li><li>Unambiguous state transitions</li><li>Exact money movement and reconciliation</li></ul><p>These models act as stress tests. They surface shortcuts that single-party systems can hide for years.</p><p>If your platform relies on informal assumptions, this is where it breaks.</p><h3>The Mental Model That Scales</h3><p><strong>Decision-time systems create promises.<br>Consequence-time systems keep those promises honest.</strong></p><p>Confuse the two and you may ship faster today, but you will pay for it in ways you cannot unwind later.</p><h3>What You Can Apply Tomorrow</h3><p>Ask yourself one question about any system you own:</p><p>Can you explain why a six-month-old irreversible decision was made, using data the system captured at the time?</p><p>If the answer is no, your decision-time platform is already leaking risk.<br>If the answer depends on tribal knowledge, it will eventually fail.</p><p>This is not a fintech problem.<br>It is a platform maturity problem.</p><h3>What’s Next</h3><p>In the next article, we will go deep into <strong>P2P lending</strong>.</p><p>Not as a business model, but as a <strong>platform stress test</strong>.</p><p>P2P forces:</p><ul><li>Multi-principal ledgers</li><li>Perfect attribution</li><li>Zero tolerance for ambiguity</li></ul><p>It is the fastest way to understand whether your LOS and LMS boundaries are truly sound.</p><p><strong>If you understand P2P, you understand platform lending.</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7b552b62d2d4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[APIs Explained: Your 2025 Guide for Product Managers]]></title>
            <link>https://medium.com/@bharaththakur_91308/apis-explained-your-2025-guide-for-product-managers-3244ce3098e0?source=rss-473be74f4400------2</link>
            <guid isPermaLink="false">https://medium.com/p/3244ce3098e0</guid>
            <category><![CDATA[api]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[tech-for-beginners]]></category>
            <category><![CDATA[tech-for-non-tech]]></category>
            <category><![CDATA[beginners-guide]]></category>
            <dc:creator><![CDATA[Bharath Thakur]]></dc:creator>
            <pubDate>Mon, 13 Oct 2025 05:20:01 GMT</pubDate>
            <atom:updated>2025-10-13T05:20:01.148Z</atom:updated>
            <content:encoded><![CDATA[<p>Every time you order groceries on Swiggy Instamart, book a cab on Rapido, pay via PhonePe, or check your Zerodha portfolio, you’re using APIs. Behind every app you love, there’s a network of APIs working silently in the background.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*o0HQRohJV2-UAYuJ" /><figcaption>Photo by <a href="https://unsplash.com/@ilgmyzin?utm_source=medium&amp;utm_medium=referral">ilgmyzin</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Think of APIs as the digital post office system. Just like India Post has standardized protocols for sending and receiving parcels — addressing formats, tracking numbers, delivery confirmations — APIs have standardized ways for applications to send and receive data. You don’t need to know how the postal network routes your package from Mumbai to Bangalore; you just need to know the address format and tracking system. Similarly, you don’t need to understand a company’s entire database architecture; you just need to know their API endpoints.</p><p>In 2025, the API economy is worth over $7 trillion globally, and India is a massive contributor. Companies like Razorpay (valued at $7.5 billion), Nium (handling $90 billion in transactions), and Postman (founded in Bangalore, valued at $5.6 billion) have built entire businesses on APIs. Even UPI, which processed 131 billion transactions in 2024, is powered by NPCI’s API infrastructure.</p><h3>What Exactly is an API?</h3><p><strong>Application Programming Interface (API)</strong> = A standardized way for two software applications to exchange information.</p><h3>The Railway Reservation System Analogy</h3><p>Think about booking train tickets on IRCTC:</p><ul><li><strong>You</strong> = Your app (like Paytm or MakeMyTrip)</li><li><strong>The ticket counter interface</strong> = The API documentation</li><li><strong>The reservation clerk</strong> = The API itself</li><li><strong>The railway database</strong> = The server with all train data</li><li><strong>Your confirmed ticket</strong> = The data returned to your app</li></ul><p>You don’t walk into the railway headquarters and directly access their mainframe. You go through the standardized booking system. The clerk checks availability, confirms your booking, and gives you a ticket. The API works exactly like this — it provides a controlled interface to access complex systems.</p><h3>Why Not Direct Access?</h3><p>Two crucial reasons: <strong>Security</strong> and <strong>Stability</strong>.</p><p>Imagine if everyone could directly query the railway database:</p><ul><li>Crores of simultaneous queries would crash the system (like IRCTC on Tatkal booking days!)</li><li>Data could be corrupted or stolen</li><li>There’d be no way to manage load or prioritize requests</li><li>No standardization means chaos</li></ul><p>APIs act as controlled gateways with rate limits, authentication, and proper data formatting.</p><h3>Real-World API Examples (2025 Edition)</h3><h3>1. Razorpay Payment API</h3><p>When you pay for your food delivery order, the app uses Razorpay’s API:</p><pre>// Simplified Razorpay payment example<br>const Razorpay = require(&#39;razorpay&#39;);</pre><pre>const razorpay = new Razorpay({<br>  key_id: &#39;your_key_id&#39;,<br>  key_secret: &#39;your_key_secret&#39;<br>});</pre><pre>const order = await razorpay.orders.create({<br>  amount: 50000, // ₹500.00 in paise<br>  currency: &#39;INR&#39;,<br>  receipt: &#39;order_rcptid_11&#39;,<br>  payment_capture: 1<br>});</pre><pre>console.log(&#39;Order ID:&#39;, order.id);</pre><h3>2. OpenWeather API</h3><p>Weather apps use this to fetch real-time weather data:</p><pre>// Fetch weather data<br>const fetchWeather = async (city) =&gt; {<br>  const response = await fetch(<br>    `https://api.openweathermap.org/data/2.5/weather?q=${city}&amp;appid=YOUR_API_KEY`<br>  );<br>  <br>  const data = await response.json();<br>  console.log(&#39;Temperature:&#39;, data.main.temp);<br>  console.log(&#39;Humidity:&#39;, data.main.humidity);<br>  <br>  return data;<br>};</pre><h3>3. Stripe Payment API</h3><p>Global payment processing:</p><pre>// Create payment intent with Stripe<br>const stripe = require(&#39;stripe&#39;)(&#39;your_secret_key&#39;);</pre><pre>const paymentIntent = await stripe.paymentIntents.create({<br>  amount: 2000, // $20.00<br>  currency: &#39;usd&#39;,<br>  payment_method_types: [&#39;card&#39;],<br>});</pre><pre>console.log(&#39;Payment Intent ID:&#39;, paymentIntent.id);</pre><h3>Types of APIs: Your Options in 2025</h3><h3>By Access Level</h3><p><strong>🌍 Public/Open APIs</strong></p><ul><li>Free or freemium for anyone to use</li><li>Examples: India Post API, OpenWeather (used by many Indian weather apps), COVID-19 India API</li><li>Great for startups and side projects</li></ul><p><strong>🤝 Partner APIs</strong></p><ul><li>Shared between specific business partners</li><li>Example: PhonePe’s APIs for merchant partners, Swiggy’s restaurant partner APIs</li><li>Requires authentication and business agreement</li></ul><p><strong>🔒 Private/Internal APIs</strong></p><ul><li>Used within a single organization</li><li>Example: Flipkart’s internal microservices, Jio’s internal systems</li><li>Not accessible to outsiders</li></ul><p><strong>⚡ Composite APIs</strong></p><ul><li>Combine multiple API calls into one</li><li>Faster and more efficient for mobile apps (crucial for 4G/5G networks)</li><li>Example: CRED’s dashboard fetching credit cards + bills + offers in one go</li></ul><h3>API Protocols &amp; Architectures: The Complete Comparison</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/778/1*McfW_V7rpDwb1S5grs93Nw.png" /></figure><h3>When to Use What?</h3><p><strong>Use REST when:</strong></p><ul><li>Building consumer-facing apps (most common choice)</li><li>You need simple, stateless operations</li><li>Working with limited resources/small team</li><li>Need wide language/framework support</li><li>Example: Building a food delivery app like Swiggy</li></ul><p><strong>Use SOAP when:</strong></p><ul><li>Working with banks or insurance companies (they still use it!)</li><li>Need strict contracts and validation (WSDL)</li><li>Enterprise integration with legacy systems</li><li>High security requirements with built-in standards</li><li>Example: Integrating with government tax filing systems</li></ul><p><strong>Use GraphQL when:</strong></p><ul><li>Mobile apps need to minimize data transfer</li><li>Frontend teams want flexibility in data fetching</li><li>You have complex, nested data relationships</li><li>Multiple clients need different data shapes</li><li>Example: Building a super app like CRED with multiple features</li></ul><p><strong>Use gRPC when:</strong></p><ul><li>Building internal microservices</li><li>Need maximum performance (10x faster than REST)</li><li>Language-agnostic service communication</li><li>Real-time streaming is required</li><li>Example: Payment gateway internal services</li></ul><p><strong>Use WebSocket when:</strong></p><ul><li>Need bidirectional, real-time communication</li><li>Building live dashboards, chat, multiplayer games</li><li>Push notifications without polling</li><li>Low latency is critical</li><li>Example: Live cricket scores on Cricbuzz</li></ul><p><strong>Use Webhooks when:</strong></p><ul><li>Need event-driven architecture</li><li>Want to notify external systems of changes</li><li>Avoid constant polling (saves server resources)</li><li>Async processing is acceptable</li><li>Example: Sending order confirmation to logistics partner</li></ul><h3>API Architectures: Detailed Breakdown</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/785/1*4xU8eJ4fVCIdhWQiJwriFA.png" /></figure><h3>Example: REST vs GraphQL</h3><p><strong>REST API</strong> (requires multiple requests on slower networks):</p><pre>// Request 1: Get user details<br>const user = await fetch(&#39;/api/users/123&#39;);</pre><pre>// Request 2: Get user&#39;s orders<br>const orders = await fetch(&#39;/api/users/123/orders&#39;);</pre><pre>// Request 3: Get order items<br>const items = await fetch(&#39;/api/orders/456/items&#39;);</pre><pre>// Problem: 3 separate network calls</pre><p><strong>GraphQL API</strong> (single request — better for bandwidth):</p><pre>const query = `<br>  query {<br>    user(id: &quot;123&quot;) {<br>      name<br>      phoneNumber<br>      orders {<br>        orderId<br>        amount<br>        items {<br>          name<br>          price<br>        }<br>      }<br>    }<br>  }<br>`;</pre><pre>const response = await fetch(&#39;/graphql&#39;, {<br>  method: &#39;POST&#39;,<br>  headers: { &#39;Content-Type&#39;: &#39;application/json&#39; },<br>  body: JSON.stringify({ query })<br>});</pre><pre>// Single request with all data - saves bandwidth and time!</pre><h3>Key Questions Every PM Must Ask</h3><p>Before choosing or building an API, consider:</p><h3>1. Network Conditions</h3><ul><li>Will users be on 3G/4G or primarily WiFi?</li><li>How do we handle intermittent connectivity? (Common in metro commutes, certain regions)</li><li>Should we implement request queuing for offline scenarios?</li></ul><h3>2. Scale</h3><ul><li>Can we handle peak traffic? (Festival sales, live events — 10–50x normal load)</li><li>What about during major cricket matches for sports apps?</li><li>Flash sales on e-commerce platforms?</li></ul><h3>3. Data Needs</h3><ul><li>Do we need real-time updates? (stock trading, live scores)</li><li>Is periodic syncing enough? (order status checks)</li><li>How much data per request? (mobile data considerations)</li></ul><h3>4. Security &amp; Compliance</h3><ul><li>Financial compliance requirements?</li><li>Authentication methods needed?</li><li>KYC verification integration?</li><li>Data privacy regulations?</li></ul><h3>5. Regional Considerations</h3><ul><li>Multi-language support in responses?</li><li>Local numbering formats in financial APIs?</li><li>State-specific data handling? (GST rates, pin codes)</li></ul><h3>6. Cost</h3><ul><li>What’s the per-request cost?</li><li>Can we negotiate better rates at scale?</li><li>Self-hosted vs cloud-hosted options?</li></ul><h3>7. Developer Experience</h3><ul><li>Is documentation clear and comprehensive?</li><li>Are there SDKs for popular tech stacks?</li><li>What’s the support response time?</li></ul><h3>Understanding API Anatomy</h3><p>Let’s break down a typical API URL:</p><pre>https://api.razorpay.com:443/v1/orders/order_KjHDSjdnKJH123/payments<br>└─┬─┘ └──────┬──────┘ └┬┘ └┬┘ └──────┬──────┘ └───┬───┘<br>Protocol   Host      Port Version  Resource ID   Endpoint</pre><p><strong>Breaking it down:</strong></p><ul><li><strong>Protocol</strong> (https://): Secure communication (mandatory for payment APIs)</li><li><strong>Host</strong> (api.razorpay.com): Server address</li><li><strong>Port</strong> (443): Connection port (usually hidden for HTTPS)</li><li><strong>Version</strong> (v1): API version (Razorpay won&#39;t break your integration with updates)</li><li><strong>Resource</strong> (orders/ID/payments): Specific data you want</li></ul><h3>Response Formats</h3><p>Indian APIs typically return data in <strong>JSON</strong>:</p><pre>// Zerodha Kite API response example<br>{<br>  &quot;status&quot;: &quot;success&quot;,<br>  &quot;data&quot;: {<br>    &quot;tradingsymbol&quot;: &quot;RELIANCE&quot;,<br>    &quot;exchange&quot;: &quot;NSE&quot;,<br>    &quot;last_price&quot;: 2847.50,<br>    &quot;change&quot;: 23.45,<br>    &quot;change_percent&quot;: 0.83,<br>    &quot;volume&quot;: 4523678<br>  }<br>}</pre><h3>HTTP Status Codes: What Do They Mean?</h3><p>When you make an API request, you get a status code:</p><h3>Success Codes (2xx)</h3><ul><li><strong>200 OK</strong>: Payment successful, data fetched</li><li><strong>201 Created</strong>: Order placed, account created</li><li><strong>204 No Content</strong>: Logout successful (no data to return)</li></ul><h3>Client Error Codes (4xx)</h3><ul><li><strong>400 Bad Request</strong>: Invalid UPI ID format, wrong pin code</li><li><strong>401 Unauthorized</strong>: Session expired, re-login required</li><li><strong>403 Forbidden</strong>: KYC not completed, region blocked</li><li><strong>404 Not Found</strong>: Order ID doesn’t exist</li><li><strong>429 Too Many Requests</strong>: Rate limit hit (common during sales)</li></ul><h3>Server Error Codes (5xx)</h3><ul><li><strong>500 Internal Server Error</strong>: Something broke (file a support ticket)</li><li><strong>502 Bad Gateway</strong>: Payment gateway timeout</li><li><strong>503 Service Unavailable</strong>: Maintenance mode or overloaded</li></ul><h3>Example Error Handling</h3><pre>async function checkPaymentStatus(txnId) {<br>  try {<br>    const response = await fetch(`/api/payments/${txnId}/status`);<br>    <br>    // Payment verification failed<br>    if (response.status === 404) {<br>      return { <br>        status: &#39;not_found&#39;, <br>        message: &#39;Transaction not found&#39; <br>      };<br>    }<br>    <br>    // Too many status checks<br>    if (response.status === 429) {<br>      return { <br>        status: &#39;rate_limit&#39;, <br>        message: &#39;Too many requests. Please retry after 30 seconds&#39; <br>      };<br>    }<br>    <br>    // Payment gateway timeout<br>    if (response.status === 504) {<br>      return { <br>        status: &#39;timeout&#39;, <br>        message: &#39;Gateway timeout. Please check with your bank&#39; <br>      };<br>    }<br>    <br>    if (!response.ok) {<br>      throw new Error(`Payment check failed: ${response.status}`);<br>    }<br>    <br>    const data = await response.json();<br>    return { status: &#39;success&#39;, data };<br>    <br>  } catch (error) {<br>    console.error(&#39;Payment status check failed:&#39;, error);<br>    return { <br>      status: &#39;error&#39;, <br>      message: &#39;Network error. Please try again&#39; <br>    };<br>  }<br>}</pre><h3>CRUD Operations: The Core Four Actions</h3><p>Most APIs perform these four basic operations:</p><h3>JavaScript Examples</h3><pre>// CREATE - Add address for delivery<br>const addAddress = async (addressData) =&gt; {<br>  const response = await fetch(&#39;/api/user/addresses&#39;, {<br>    method: &#39;POST&#39;,<br>    headers: { <br>      &#39;Content-Type&#39;: &#39;application/json&#39;,<br>      &#39;Authorization&#39;: &#39;Bearer &#39; + authToken<br>    },<br>    body: JSON.stringify({<br>      name: addressData.name,<br>      phone: addressData.phone,<br>      pincode: addressData.pincode,<br>      address: addressData.address,<br>      city: addressData.city,<br>      state: addressData.state,<br>      addressType: &#39;home&#39; // home, office, other<br>    })<br>  });<br>  return response.json();<br>};</pre><pre>// READ - Check order status<br>const checkOrderStatus = async (orderId) =&gt; {<br>  const response = await fetch(`/api/orders/${orderId}`);<br>  const data = await response.json();<br>  <br>  console.log(&#39;Order ID:&#39;, data.orderId);<br>  console.log(&#39;Status:&#39;, data.status); // Confirmed, Processing, Shipped, Delivered<br>  console.log(&#39;Expected Delivery:&#39;, data.estimatedDelivery);<br>  <br>  return data;<br>};</pre><pre>// UPDATE - Modify delivery address<br>const updateAddress = async (addressId, newData) =&gt; {<br>  const response = await fetch(`/api/addresses/${addressId}`, {<br>    method: &#39;PATCH&#39;,<br>    headers: { <br>      &#39;Content-Type&#39;: &#39;application/json&#39; <br>    },<br>    body: JSON.stringify({<br>      phone: newData.phone,<br>      address: newData.address<br>    })<br>  });<br>  return response.json();<br>};</pre><pre>// DELETE - Cancel order<br>const cancelOrder = async (orderId, reason) =&gt; {<br>  const response = await fetch(`/api/orders/${orderId}`, {<br>    method: &#39;DELETE&#39;,<br>    headers: { &#39;Content-Type&#39;: &#39;application/json&#39; },<br>    body: JSON.stringify({<br>      cancellationReason: reason,<br>      refundMode: &#39;source&#39; // source, wallet, bank<br>    })<br>  });<br>  return response.json();<br>};</pre><p><strong>PUT vs PATCH:</strong></p><ul><li><strong>PUT</strong>: Replace entire address (all fields required)</li><li><strong>PATCH</strong>: Update just phone number (other fields stay same)</li></ul><pre>// PATCH - Update only phone number<br>await fetch(&#39;/api/addresses/addr_123&#39;, {<br>  method: &#39;PATCH&#39;,<br>  body: JSON.stringify({ phone: &#39;+919876543210&#39; })<br>});</pre><pre>// PUT - Replace entire address<br>await fetch(&#39;/api/addresses/addr_123&#39;, {<br>  method: &#39;PUT&#39;,<br>  body: JSON.stringify({<br>    name: &#39;Rahul Sharma&#39;,<br>    phone: &#39;+919876543210&#39;,<br>    pincode: &#39;560001&#39;,<br>    address: &#39;Flat 301, Brigade Towers&#39;,<br>    city: &#39;Bangalore&#39;,<br>    state: &#39;Karnataka&#39;<br>  })<br>});</pre><h3>Metrics Every PM Should Track</h3><h3>Business Metrics</h3><ul><li><strong>API Adoption Rate</strong>: How many merchants/developers using your API?</li><li><strong>Time to First Transaction</strong>: How fast can a developer integrate and go live?</li><li><strong>Transaction Success Rate</strong>: Critical for payment/fintech APIs (aim for &gt;99%)</li><li><strong>GMV via API</strong>: Total transaction value flowing through APIs</li></ul><h3>Performance Metrics</h3><ul><li><strong>Latency</strong>: Response time across regions</li><li><strong>Uptime</strong>: Target 99.9%+ (especially during peak traffic)</li><li><strong>Error Rate</strong>: Keep below 0.5% for production</li><li><strong>Peak Load Handling</strong>: Can you handle high-traffic events?</li></ul><h3>Developer Experience Metrics</h3><ul><li><strong>Documentation Bounce Rate</strong>: Are developers finding answers?</li><li><strong>Support Ticket Volume</strong>: What are common integration issues?</li><li><strong>SDK Downloads</strong>: Node.js, Python, PHP, Java adoption</li><li><strong>Postman Collection Downloads</strong>: Are developers using your examples?</li></ul><h3>Domain-Specific Metrics</h3><ul><li><strong>Regional Performance</strong>: How fast is API across different locations?</li><li><strong>Festival/Event Load Handling</strong>: Traffic spikes during major events</li><li><strong>Payment Method Mix</strong>: Different payment methods via API</li><li><strong>Transaction Success Rate</strong>: Critical for payment/fintech APIs (aim for &gt;99%)</li></ul><h3>Essential Tools for API-Driven PMs (2025)</h3><h3>1. API Testing &amp; Development</h3><ul><li><strong>Postman</strong>: Industry standard (built in Bangalore!)</li><li><strong>Thunder Client</strong>: Lightweight VS Code extension</li><li><strong>Hoppscotch</strong>: Open-source alternative (built by Indian developer)</li></ul><h3>2. API Documentation</h3><ul><li><strong>Postman Docs</strong>: Auto-generated from collections</li><li><strong>Readme.io</strong>: Used by Razorpay, Cashfree</li><li><strong>GitBook</strong>: Used by many Indian startups</li></ul><h3>3. API Monitoring</h3><ul><li><strong>Sentry</strong>: Error tracking</li><li><strong>DataDog</strong>: Enterprise monitoring</li><li><strong>Better Stack</strong>: Modern monitoring solution</li></ul><h3>4. API Management</h3><ul><li><strong>AWS API Gateway</strong>: Scalable cloud solution</li><li><strong>Kong</strong>: Open-source gateway</li><li><strong>Azure API Management</strong>: Microsoft cloud platform</li></ul><h3>5. API Analytics</h3><ul><li><strong>Mixpanel</strong>: Product analytics</li><li><strong>Amplitude</strong>: User behavior analytics</li><li><strong>Google Analytics 4</strong>: Free tier available</li></ul><h3>6. Payment API Testing</h3><ul><li><strong>Razorpay Test Mode</strong>: Sandbox for testing</li><li><strong>Stripe Test Environment</strong>: Pre-production testing</li><li><strong>PayPal Sandbox</strong>: Mock transactions</li></ul><h3>7. Collaboration Tools</h3><ul><li><strong>Notion</strong>: Documentation and knowledge management</li><li><strong>Linear</strong>: Modern project management</li><li><strong>Slack</strong>: Team communication</li></ul><h3>8. Government &amp; Verification APIs</h3><ul><li><strong>Tax Verification APIs</strong>: GST and tax compliance</li><li><strong>Identity Verification</strong>: KYC and document verification</li><li><strong>Digital Document Storage</strong>: Secure document APIs</li><li><strong>Financial Data APIs</strong>: Banking and transaction verification</li></ul><h3>Final Thoughts</h3><p>APIs are no longer just technical infrastructure — they’re business strategy. Every successful tech company has built its competitive advantage with APIs. Razorpay democratized payments, Postman made API development accessible globally, and various payment systems showed how to do digital transactions at scale.</p><p>As a PM, understanding APIs helps you:</p><ul><li>Make better build vs buy decisions</li><li>Reduce time to market (integrate instead of building)</li><li>Scale efficiently (leverage existing infrastructure)</li><li>Enter new markets faster (standardized APIs, global platforms)</li></ul><p>The best APIs are:</p><ul><li><strong>Simple</strong> to integrate (good docs, SDKs)</li><li><strong>Reliable</strong> (99.9%+ uptime, especially during peak periods)</li><li><strong>Fast</strong> (optimized for various network conditions)</li><li><strong>Affordable</strong> (pricing that works at scale)</li><li><strong>Well-supported</strong> (responsive support, clear documentation)</li></ul><p>Whether you’re building a fintech app, a D2C brand, or a SaaS product, mastering APIs will make you a significantly more effective PM.</p><p><strong>The API economy is just getting started. Get in now.</strong></p><p><em>What APIs are you building with? Working on anything interesting? Let me know in the comments!</em></p><p><em>If this helped you understand APIs better, share it with your PM network. Subscribe for more deep dives on product management in India.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3244ce3098e0" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>