Atul Krishna

Most CS engineers in India are effectively useless for Day 1 impact. — One of the biggest reasons we started Polaris School of Technology was to solve a problem that we ourselves faced internally at Classplus Back in 2018, we were building SaaS, and needed software devs, obviously. Easiest call was to go for fresher CS grads - they are young energy + agile learning + early stage cos can afford only them But the shock of our lives, almost no fresher we hired could push live code on day 1. And the simplest explanation - "we didn’t learn how to do this in college."   And on asking why did they not - "Nobody exactly told us how to do so."    And max kids just prepared some puzzles on internet to crack campus placements. Funny max! — And this was not just with us, almost every tech founder we know was facing the same.    Resulting in two options: One, abandon and only go for senior talent, but non-feasible for the costs + org structure involved Two, train and upskill the fresh talent and make them industry-relevant over 6-9 months and see them contribute meaningfully post that. Each, every, and 100% of the founders I know in life opted for the second one, and the first jobs for most software engineers became the playground where they learnt all their coding skills. The bar was so low to be honest, that anyone who had done a couple of good internships / built some functional products / written actual code during college times was getting 2x CTCs than others. The sad part: All of this is still true in most cases. Most CS grads actually learn real coding / software development on the go, gradually rising through the skills + payscale. And this pushes back everyone - startups lose time in building products / engineers lose some of their sharpest years / the ecosystem moves slower than it should. — And that’s why we built Polaris with a lot of heart. Not just coz we find it worthy, but also because the industry needs this. Young CS grads who interact with real companies while studying, build live products, do a few internships, network with existing devs, participate in a few hackathons, and have enough confidence / clarity before they step in full-time. No training wheels. No "6-month ramp-up." Just Day 1 ready :)

246

12 Comments

We just shipped one of our largest enterprise GenAI deployments at MerakiPeople. Here's what it actually took to go live: - Syncing thousands of leads and hundreds of thousands of activity signals across two enterprise platforms. - Building audio pipelines that parse Hindi-English-Hinglish code-switching mid-sentence. - Creating evidence-backed scoring engines — not vibes, actual timestamped quotes from conversations. - Designing #hyperpersonalized nurture that considers someone's city, career stage, family situation, and language preference. Not "Hi {First_Name}" personalization. Real, contextual, human-sounding communication. At scale. And then the hardest part — building validation layers that catch when the AI hallucinates instead of quietly faking an analysis. Because here's what I've learned: GenAI at scale is 20% model, 80% plumbing. The separation between client-owned strategy (what to say, when) and product-level intelligence (how to personalize for each individual) — that's the real architecture decision most people get wrong. But here's what made my week. When I sat with the sales team and asked — "Will this actually be useful?" — the response was instant: absolutely. One counselor told me he's already using ChatGPT to help with his conversations. That's GenAI positivity already living on the floor. The difference? With MerakiPeople, he doesn't need to context-switch to a separate tool. The intelligence is seamless, built around his leads, his conversations, his pipeline. Not generic prompts — contextual, continuous, and compounding with every call. We don't replace salespeople. We liberate them to be human. Sanjoy Paul __________________________________________________________________________ If you're running a B2C sales floor — EdTech, FinTech, InsurTech, HealthTech — and your team is doing 100+ calls a day with zero learning loops, zero personalized follow-ups, and zero visibility into what's actually working on those calls? Let's talk. DM me or comment below. We're onboarding a visioanry set of partners who want to co-build the future of Human + AI sales. #GenAI #SalesEnablement #AIatScale #MerakiPeople #BuildInPublic #B2CSales

14

The recent change in GST rates across India is a textbook example of why modular, well structured code is a necessity, not a luxury. Imagine this: Marketplaces like Amazon, Flipkart, or even offline billing systems suddenly had to update the recent GST rates across thousands of SKUs, multiple categories, and countless seller dashboards overnight. If their systems were written as one large, monolithic block of code where tax logic is scattered everywhere, updating a single rate would have required: • Touching multiple codebases • Risking regressions in unrelated modules • Testing the entire application end to end The result? Delays, bugs, and chaos at scale. But because most modern platforms follow modular architectures where tax computation, order management, payment gateways, and invoices are handled by distinct, decoupled modules, updating GST rates is as simple as: • Changing a configuration or tax-service module or a value in the database. • Running focused tests on that module. • Deploying quickly with minimal ripple effects. This is why you probably saw marketplaces adapt to the new GST rates almost instantly. They could change one module without breaking the rest of the system. Takeaway: Modular code is like having separate circuit breakers in your house. If one needs an upgrade, you don’t shut down the entire grid. So, the next time you’re tempted to skip modularity “because it takes longer upfront,” remember: When the next GST-like change hits, speed, safety, and scalability will only come from a modular design. #SoftwareEngineering #ModularCode #CleanArchitecture #GST

25

Over the past few months, I’ve been working on a challenge - building a scalable payment ecosystem where we can plug & play any number of payment partners like PayU, PhonePe, Cashfree Payments, Razorpay and more. The result? A modular architecture that separates database logic from partner-specific implementations, making the system robust, scalable, and super easy to extend. Wrote a detailed post about the architecture, patterns, and my thought process behind it. Read it here 👇 https://lnkd.in/gh8HYbSg #payments #systemDesign #scalable #architecture #pattern

17

1 Comment

ChatGPT is executing a classic platform strategy, aiming to become the "prompt-based operating system" 🤯 Many vendors have tried the super-app concept (PhonePe, Tata Neu), but what ChatGPT is attempting is off the charts. The Vision: 📱 "Ask / Browse / Learn anything within ChatGPT" - Search / Browser 🛒 "Buy anything within ChatGPT" - Commerce / Fashion ⚡ "Prompt / Talk-to to get things done" - Productivity 🔗 "Run your business within ChatGPT" - CRM / Sales At their recent DevDay, OpenAI dropped the real game-changer that makes all this possible. 📲 Apps Inside ChatGPT: Forget app stores. Developers can now build their apps directly inside ChatGPT. Imagine using Canva, booking a flight, or listening to Spotify just by talking to ChatGPT. Major partners like Spotify, Figma and Coursera are already on board. What does this actually mean? ChatGPT is collapsing the roles of an operating system, an app store, and a personal assistant into one single, conversational platform. With over 800 million weekly users, the magnetic pull for developers to join is undeniable 🚀 Competition? Nowhere near when it comes to consumer market share. Gemini, Anthropic, Copilot and Perplexity will get their shares, but ChatGPT is leading the race. The Big Question: What happens when a single vendor essentially owns the "everything app"? 🤔 What's Next (my prediction)? ⏭️ Hardware that integrates with the ChatGPT ecosystem of apps. Probably a fork of Android that would be more 'prompt/talk-to' focused over the traditional touchscreen interface. Exciting times ahead! 🔮 I won't be surprised if Android smartphones sold next year advertise "Supports ChatGPT 2.0." Why 2.0? I have a strong feeling that ChatGPT apps will definitely require breaking out of the standard Android application sandbox.

7

Last 48 hours = pure debugging madness. Our app was randomly failing. Auth breaking. API calls timing out. Edge functions not responding. At first, we thought: Supabase outage? SSL issue? Region latency problem? Nope. Turns out : if a user is on Jio all *.supabase.co domains are resolving to a Jio-owned IP (49.44.79.236) instead of AWS. Which basically means: Your Supabase backend is unreachable. Not slow. Not degraded. Just… dead. What’s wild: supabase.com loads fine App works perfectly on Airtel / Vi Works instantly on VPN Only supabase.co (project API domains) are affected nslookup → wrong IP curl → timeout traceroute → dies inside Reliance network This isn’t a project bug. This isn’t a code bug. This isn’t even a Supabase bug. It looks like DNS-level blocking inside Jio’s network. If you're building for India, this is scary. Because you won’t see a crash. You’ll just see: Drop in signups OTP failures Random API errors Users ghosting mid-flow And you’ll assume it’s your funnel. Given how big Jio’s user base is, this could impact a massive chunk of Indian traffic without founders even realizing it. and to save your time I believe that Jio appears to be DNS-poisoning *.supabase.co to a sinkhole IP and Changing DNS to 1.1.1.1 or 8.8.8.8 alone did not help us. Guys please suggest Temporary workarounds: Except using VPN as cloudflare and Proton vpn didnot work as well for me. If you’re using Supabase and targeting Indian users, please test your app specifically on Jio broadband + Jio mobile data. This kind of infra issue is the worst kind ,silent and invisible. Would love to know if other builders are seeing the same thing. #Supabase #StartupIndia #DevLife #BuildInPublic #SaaS #IndiaTech

15

When Supabase suddenly became inaccessible across parts of India, it was not because of downtime, not a bug, not scaling issues. It was simply a reminder that in today’s world, even core infrastructure can disappear overnight due to external factors. It made me think about how we build products now. Most of us move fast by stacking multiple external services together. It works great until one layer stops working and you realise how dependent everything has become. A few simple takeaways from this: Speed is important, but ownership of core data is more important. Never let a single third party become your only source of truth. Hybrid setups are safer than fully managed or fully self-hosted extremes. Always design systems assuming outages will happen someday. This is not about avoiding modern tools. They are still the best way to build quickly. It is just about being a little more intentional about where we place our biggest dependencies. #supabase #india #ai

9

1 Comment

You can now order biryani on ChatGPT Not a demo. Not a concept. Swiggy Instamart has MCP live with 40,000+ products exposed to ChatGPT and Claude, Zomato too. COD only for now, but the direction is clear. The narrative is that AI will disintermediate platforms. That ChatGPT will replace Swiggy. That Perplexity AI will replace Flipkart. I spent the last few weeks mapping what's actually happening. The reality is way more interesting than the doomsday takes. Every major Indian e-commerce platform is deploying GenAI across the full stack: - Eternal (Zomato + Blinkit) has a strategic OpenAI partnership spanning 5 verticals - Meesho deployed India's first GenAI voice bot handling millions of support calls in Hindi, Tamil, Telugu - Flipkart rebuilt product search from keyword matching to neural embeddings - Swiggy became the first quick-commerce service globally to adopt MCP - Amazon India is running Project Suraksha to detect counterfeit listings at scale AI apps will become a demand channel (like Google), not a disintermediator. Because e-commerce moats are built on supply chain orchestration, not discovery. Dark stores, cold chains, fleet management, restaurant onboarding. ChatGPT can't do that. TimeSpent ≠ Trust or Transactions. I mapped 22+ GenAI initiatives across 5 companies, built a defensibility spectrum, and broke down the MCP case. Full deep dive here: https://lnkd.in/gHSN4x44

10

🚨 Frontend Developers, Listen Up! If you’re still not exploring AI integration in your frontend apps, you’re missing the biggest shift in web development since SPAs became a thing. ➡️ AI isn’t just for backend or data teams anymore. • Intelligent UI interactions • Personalized user journeys • Context-aware components • Agentic apps that think and act 📌 The question isn’t “Will AI change frontend?” — it already is. The question is: “Will you be one of the devs building that future or catching up to it?” 🔥 Start small. Add an AI-powered feature. Experiment with context-aware UIs. But whatever you do, don’t ignore it. 👉 Want to see how AI and frontend can come together? Join us at the NG DELHI meetups where we’re exploring exactly that with live talks, demos, and hands-on sessions. #frontend #webdevelopment #ai #angular #react #ngdelhi #futureofweb

16

Zepto’s “Swap & Save” – Game-Changer or Tightrope Walk? Zepto recently introduced a feature called “Swap & Save”, where customers are offered lower-priced alternatives to items in their cart. As a product manager, this move got me thinking deeply about the long-term implications for both customers and brands. Here’s my take: For Customers: • Short-term win: Users are likely to try cheaper alternatives at least once — especially in commoditized categories. • Long-term risk: If the swapped product doesn’t meet expectations, users revert to their original choice. Worse, they may lose trust in the platform’s recommendations. For Brands: • Pricing pressure: Brands may feel compelled to lower prices to avoid getting “swapped out.” • Power imbalance: If Zepto starts favoring select or private-label brands in swaps, it could trigger fair trade concerns. • Incentive to play along: Some may offer deeper discounts or pay for visibility — but is that sustainable or ethical? Platform Dynamics: • Zepto is becoming a gatekeeper of choice, influencing not just logistics, but purchase decisions. • If done right, this feature can help optimize inventory, reduce costs, and improve margins. • But the fine line between smart nudges and manipulative practices must be respected. In product, what works in the short term can backfire long term if we don’t balance user trust, partner fairness, and business goals. Would love to hear thoughts from others in product, retail, and e-commerce. Is “Swap & Save” a customer-first feature — or a slippery slope? #ProductManagement #Ecommerce #Zepto #PlatformStrategy #CustomerExperience #BrandStrategy

2

Back in August, I interviewed with PhonePe, and that experience left me with a valuable lesson. The feedback I received was simple yet powerful: “You did well, but you didn’t think about the requirements.” And they were absolutely right. I jumped straight into designing a scalable system discussing databases, APIs, caching, and all the fancy stuff. But I never paused to ask “What if this system was meant to be small-scale?” That one question could have completely changed my approach. It reminded me that system design isn’t about complexity, it’s about context. Before scaling, you first need to understand what problem you’re solving and for whom. I’ve written a short blog sharing my interview experience what went well, what I missed, and all about other rounds. 👉 Link in the comments.

115

3 Comments

💰 Groww is not just adding users — it is upgrading the quality of the user base. The ARC’s breakdown shows the “Two Indias” inside Groww: 🟩 Aspirational India This is still the bulk of users + revenue. But their AARPU (average revenue per user) is almost plateauing near ₹8.2K annually. 🟪 Affluent India This small but fast growing segment is where the monetisation engine is firing. Affluent user assets on Groww grew at 163% CAGR since FY22. 📌 By FY25, affluent users controlled 31% of Groww’s total assets even though they are a much smaller base. That’s the big story. 🚀 Groww is quietly becoming a wealth platform — not just a broking platform. And as affluent share expands toward 33%+ (the FY26 projection), operating leverage goes vertical. The market obsession is always “DAUs & MAUs”. But real internet power is: 💵 more assets per customer ⏳ longer retention 📈 higher monetisation naturally This is exactly where Groww is heading. https://lnkd.in/gXp6GJ5F

1

A few years ago, Byju’s and PhysicsWallah were both selling the same thing: online education. But their paths couldn’t have been more different. Byju’s scaled by renting attention. Ads everywhere. Celebrity endorsements. IPL campaigns. It worked for a while, but CAC kept rising. At its peak, Byju’s was spending close to Rs 3,000 - Rs 4,000 to acquire a customer for a course that cost only Rs 8,000 - Rs 10,000. That meant 30–40% of revenue was going straight into ads. When the funding slowed, the P&L cracked. PhysicsWallah went the other way. Alakh Pandey taught on YouTube for free, building trust every single day. His videos have crossed 3 billion views. By the time he launched paid courses at Rs 999, lakhs of students converted almost instantly. The customer acquisition cost was close to zero because the attention was already owned. That difference in unit economics is why one company is struggling while the other became successful with healthy margins. With upcoming $5 bn IPO, PhysicsWallah is also using some of those growth tactics now, so we will see how it plays out in long run. Now think about this in the AI era. Ads will still work, but they will get more crowded and expensive. Because AI will make running them ridiculously easy. Anyone with ChatGPT and other AI tools can spin up 100 ad creatives, test them in real-time, and optimize bids automatically. The barrier to run ads is falling to zero, which means competition (and ad costs) will rise. Renting attention will become a money pit. Owning attention, on the other hand, will compound faster. AI is turning every individual into a media company. With one camera and the right AI tools, you can now create 100+ videos a month instead of 10. Auto subtitles and translations mean your content reaches audiences in 10+ languages simultaneously. Distribution is no longer linear. Here is how the numbers play out for a simple Rs 999 digital course aiming for Rs 10+ cr revenue. - If you rely only on ads, you might spend Rs 500 per student on acquisition. With 1 lakh students, that’s Rs 5 crore gone on ads. Margins shrink to about 30-40% with other added costs. - If you have a YouTube channel with 500,000 subscribers, and just 20% of your buyers come organically, your blended CAC drops to ~₹300. Now margins jump to 55-60%. - If one reel goes viral and brings you 50,000 buyers at almost zero cost, your margin curve bends dramatically upward. The P&L flips from fragile to resilient. This is the quiet shift AI is driving. Renting attention will only get costlier. Owning attention will become exponentially more valuable. In the next decade, the biggest winners won’t be those who bought the most ads. They will be the ones who earned trust, built audiences, and then used AI to scale that trust at near zero marginal cost.

16

‘After 2014, there was suddenly an enormous demand to understand the right-wing ecosystem.’ Read more from Abhishek Choudhary on Vajpayee and BELIEVER’S DILEMMA in his interview with Kavitha Iyer for Article 14: https://lnkd.in/dQQwrczc 

25

1 Comment

The Razorpay MCP Server is officially live! Razorpay becomes India’s first payment gateway with an Official MCP (Model Context Protocol) Server, designed for an AI-first world. 🔗 Connect to any AI system, whether it's a chat, an app, or a fully autonomous system - you can embed payments within it. For example: Type "Send ₹500 payment link to Neha", and a payment link is generated, shared on WhatsApp, and completed, all within seconds. Here's how to try it: 1. Install Razorpay MCP server 2. ⁠Configure MCP client such as Claude Desktop or VS Code or Cursor 3. ⁠Unlock the power of Razorpay inside your favourite AI assistant (Such as Claude) Access the full documentation on GitHub: https://lnkd.in/gNSvjpEg

5,100

121 Comments

Speed wins in AI, and infrastructure should never be the bottleneck 🚨 When Techdome set out to accelerate AI-led product delivery across India, the UAE, and the US, they needed a Kubernetes platform that matched their automation-first engineering culture. Traditional cloud providers were slowing them down with high costs, complex provisioning, and over-engineered setups. By adopting Civo Kubernetes, Techdome reduced infrastructure costs, simplified operations, and provisioned production-ready clusters in minutes. The result is faster go-to-market timelines and improved ROI for both Techdome and its customers. Read the full case study to see how Techdome accelerated AI delivery with Civo: https://civo.io/4sUid34

1

In India, Blinkit handles approximately 1.7 million deliveries per day, often with a delivery window of 10 minutes. Here is a quick look at their tech stack, which made things possible for them. Redis is being used as an in-memory datastore for distributed locking, which is enabling optimistic locking at scale. https://lnkd.in/g6QrYJw5 #Redis #Blinkit #Ecommerce #DistributedLocking #Realtimesystems

24

Cloudflare is down AGAIN! From Zerodha, Groww, to Canva, Zoom, Discord and SaaS platforms, users were hit with 500 Internal Server Errors and Bad Gateway screens across the board. When a company like Cloudflare goes down, it isn’t “a website issue.” - It’s a global infrastructure event. - Impact on trade during active market hours. - Zerodha even had to route users to WhatsApp for order management - Even DownDetector, the site we all rush to check if the internet is down, was itself down. Because Cloudflare isn’t just a CDN. It’s the backbone layer for: - API routing - DNS - DDoS protection - Load balancing - Reverse proxy - Global traffic acceleration Take that layer out, even briefly, and entire businesses stall. ⚠️ What makes this outage significant is not just the scale, but the timing: - Trading platforms, payment gateways, and high-traffic consumer apps all experienced visible downtime. And here’s the uncomfortable truth: 👉 When you’re dependent on a single network edge provider, your uptime becomes their uptime. This outage is a reminder for engineering teams to think beyond the happy path: - Do you have multi-CDN fallback? - Can your critical APIs bypass Cloudflare if needed? - Is your DNS redundant across providers? - Can your app degrade gracefully without edge caching? Cloudflare will fix the issue. But the question is, will companies rethink their dependency layers after this? - The outage wasn’t just a technical incident. - It was a resilience test for the modern internet. #Cloudflare #CloudflareDown #TechNews #StockMarket #Zerodha #DevOps #InternetOutage

2,292

69 Comments

Big milestone in the tech world: Anthropic - founded just five years ago, is now valued at about USD 380 billion after a major funding round. That valuation currently tops the combined market value of several of India’s largest IT firms including Tata Consultancy Services, Infosys, Wipro and HCL Technologies. Here’s a reminder of how fast AI is reshaping the global tech landscape. This is not just about size, but about growth models, innovation and future opportunities in enterprise tech. #AI #Anthropic #TechValuations #IndianIT #Innovation

2