Pankaj Kumar

It was 4:36 PM at the Amazon India office when Ajay, a senior SDE, saw a ping from the new grad, Riya. “Hey Ajay, the order tracking service is failing after my refactor. Could you help me debug?” Ajay’s first instinct was that tiny sting of “my code” pride. He’d written the original order tracking service during the Prime Day rush. He’d lost weekends getting every edge case right. Now, someone was making changes. He remembered the old days at Infosys, too, how his seniors would scoff anytime a “newbie” wanted to touch legacy code. He nearly typed out: “Why were you even touching that module?!!” But he stopped himself.  Instead, he hopped on the call. Riya already looked nervous on the Zoom. Ajay didn’t scold. He asked a few clarifying questions, let her walk him through the bug, and together, they found a missing check that broke the flow for certain orders. 40 mins later, the fix was deployed and the service was restored. Before signing off, Riya said, almost whispering, “Sorry if I messed up your code. I thought it would make the new feature easier.” Ajay smiled, remembering all the times his own code was refactored at Amazon, and how hard it was not to take it personally. So he told her, “If nobody ever changed the code, Amazon would still be using scripts from 2010. Every improvement starts with a little discomfort.” Next day, they sat together, cleaned up the changes, and even improved the onboarding docs for the next developer. Later, in the team Slack, Ajay posted: “If someone is confident enough to touch your code, it’s a win. Refactoring is how we ship at scale. Don’t get too attached.” If you’re a software engineer, remember: You will always feel a bit possessive about your code. But the real impact is letting go, so others can make it even better. That’s how the best companies evolve. And how engineers grow, too.

5,062

141 Comments

Swiggy search returns Croissant results when you search for Prashant 🤯 I just published a video explaining how it is implemented. This isn't regular text search, phonetic search, or even semantic search. Very likely it is implemented using synonymic query expansion. In the video, I covered all three types of search - how they work, how they differ, and most importantly, how synonymic query expansion is implemented. Link in the comments. btw, I used to cover this in my system design course as well, now I need to figure out something new to cover in the next batch.

997

44 Comments

🚀 सहज जीवन: probably the first open-source, collaborative, public, and live Marathi book, built transparently on GitHub. 🚀 The first part is a translation of Leo Babauta's "The Effortless Life" in Marathi, and the second part is open for contributions. I’ve already added a couple of chapters there myself. The idea is simple: anyone with useful insights about leading a good life can join and contribute. Contributions are voluntary, no money involved. 💡 Eligibility criteria for contributors: - Fluency in reading and writing Marathi - Basic GitHub literacy (fork, commit, PR). LaTeX expertise is not required - And yes, you must have read almost all of PuLa. That’s non-negotiable 😉 You can either refine the translation of the pending chapters or add fresh chapters of your own. Just fork the repo, add your changes, and send a PR. You can also  your add your name as co-author, thus build your own version of the book into a PDF, print it for yourself, and even publish it for wider circulation. I’m planning to self-publish a slightly updated version via Notion Press, which will naturally come with some costs. Feel free to do the same, the PDF is freely available in the repo. If you want to add a chapter, just place it at the end and submit a PR (Pull Request). If you are not comfortable with LaTeX or GitHub, you can share me the content via Google-Docs or email, I will add that on your behalf. 👉 If you’re interested in this whole “effort,” drop a comment below or message me your Gmail ID. I’ll schedule a short online meet or create a chat group to align next steps. IMO, this is not just a book, it’s a movement: bringing open-source culture to Marathi literature. Imagine being part of a collective that publishes a free, evolving, community-owned book! 📎 Repo name: yogeshhk/SahajJeevan (... on GitHub). 📑 Attaching the partial draft PDF here for a quick taste. (It’s a very rough draft, just to give you the idea.) जय महाराष्ट्र ✨

110

12 Comments

At Zomato, the Home page and Search are powered by in-house, self-managed Solr clusters. Over time, the team acquired the expertise needed to handle the nuances of self-management. Despite that, we still experienced some unexpected downtimes — a few of which were relatively unknown to the broader community, probably because we were using Solr at a significantly high scale and had modelled our use cases quite differently. I’ve discussed one such downtime in detail in this blog: https://lnkd.in/gDJuBU_P While we kept improving the availability of the systems, we also built fallback disaster recovery mechanisms to handle downtimes. I had written about one of those in a previous post — it talked about serving the Zomato Home page using DynamoDB when the main flow was down. But that was only a partial recovery mechanism. Needless to say, customers rely heavily on Search to find their favourite dishes and restaurants. So we had to build a fallback mechanism for Search as well — since DynamoDB doesn’t support full-text search. Customers typically search either for a Dish/Cuisine or a Restaurant. While restaurant search is hyperlocal, Dish/Cuisine search is more universal. Interestingly, the majority of searches are for dishes — and even more interestingly, ~99% of those dish searches are confined to fewer than 1000 dishes globally. We looked at data for the top ~1000 dishes and discovered popular keywords customers use to search for them — e.g., sezwan for Schezwan, panir for Paneer, etc. We created an in-memory Trie of these strings, populated at container boot time. So, if the Solr index is unavailable, we can use this Trie to return matching dishes for the input keyword. This automatically handles phonetic matches and spelling mistakes for most user inputs. We could’ve gone further and implemented 1–2 edit distance matching, but the impact would’ve been negligible — the keyword data was already sufficient to serve the majority of real customer queries. All in all, this was a sub-10MB, almost no-cost in-memory data structure, capable of serving Dish/Cuisine searches with sub-millisecond latency on the server side. The result ranking was based on the universal popularity of dishes. While the results aren’t personalised and don’t include restaurant data, they still allow customers to search for their favourite dishes during downtimes — and in such cases, availability takes precedence over consistency. #datastructure #solr #trie #search

430

20 Comments

The Avg. Comp for Meta E5 Level in India is over 75 Lakhs+, including bonuses and stocks, it goes over 1Cr+, and this is just at the Sr. Software Engineer level. That is what Meta puts on the table if you can clear the process and it’s not too complicated:  - There are 2 DSA Rounds - There is 1 System Design Round - There is 1 Managerial / Behavioural Round Note: They’re also testing a new AI-based round, where you’re free to use AI but have to deliver a full functioning app in limited time. But they will grill you beyond what you can imagine, and that’s why you need bulletproof fundamentals, especially with DSA and System Design. If I were targeting Meta E5, these are the resources I would use to study up on my fundamentals: Before we get into resources, if you’re preparing for system design/coding interviews, our mock interview tool. You can use it for free here: https://lnkd.in/gpCn7t2T Plus at layrs.me , you can already study and improve your system design skills, you get: – 60+ interactive problems – AI-based interview to test yourself – Option to add your own problem – Real-time, first-principles feedback – A community that shows up for each other 1. APIs: https://lnkd.in/ezwnCGqS 2. API Gateways: https://lnkd.in/eqNrc77q 3. JWTs: https://lnkd.in/eAnfnzm7 4. Webhooks: https://lnkd.in/eF6gPzVJ 5. tRPC, gRPC, GraphQL, or REST: when to use what?: https://lnkd.in/eydTuVj3 6. Load Balancing: https://lnkd.in/ewTeu-58 7. Proxy vs Reverse Proxy: https://lnkd.in/enEy9QYD 8. CAP Theorem: https://lnkd.in/eePkq2kJ 10. Sharding: https://lnkd.in/e8Gyr4G2 (From 0:30 to 1:23:40) 11. Caching: https://lnkd.in/e8Gyr4G2 (Go to 1:39:08 ) 12. Scaling: https://lnkd.in/e8Gyr4G2 ( Go to 2:25:15 ) 13. Availability: https://lnkd.in/eEQ5MAnC 14. Services in System Design:https://lnkd.in/exyDGmSe 15. Databases in System Design: https://lnkd.in/eifbKsr6 16. Data Sharding and Partioning: https://lnkd.in/eVhzCnW5 17. Sync Vs Async: https://lnkd.in/ekrADFHy 18. REST: https://lnkd.in/eY2ACHFC 19. Algorithms in Distributed Systems: https://lnkd.in/eXiJ9_GV 20. SQL vs NoSQL: https://lnkd.in/entah3zc 21. ACID Transactions: https://lnkd.in/etXk_wa4 22. Consistent Hashing: https://lnkd.in/eYgXNHz4 23. CDC: https://lnkd.in/efeP3fXP 24. Caching: https://lnkd.in/eqDfvdvB 25. Caching Strategies: https://lnkd.in/eqFTdS_v 26. Cache Eviction Policies: https://lnkd.in/ewB5MZ7z 27. CDN: https://lnkd.in/eCSccEkz 28. Rate Limiting Algorithms: https://lnkd.in/etby2w5C 29. Message Queues: https://lnkd.in/eKQWVxqw 30. Bloom Filters: https://lnkd.in/eq6hN3Nn 31. Idempotency: https://lnkd.in/e-sB7a3w 32. Concurrency vs Parallelism: https://lnkd.in/eRpCq8KQ 33. Long Polling vs WebSockets: https://lnkd.in/eYZnk-93 34. Stateful vs Stateless Architecture: https://lnkd.in/egXhAmY4 35. Batch vs Stream Processing: https://lnkd.in/ez5v_suJ

988

13 Comments

🚀 Introducing BharatMLStack – Meesho’s Open-Source ML Platform 🇮🇳   We’re thrilled to announce the first release of BharatMLStack, a production-grade ML platform built at Meesho and now open-sourced for the community ❤️ ! We’re launching with our Online Feature Store, battle-tested at 1M+ QPS, serving ~100 IDs per query while consistently maintaining <10ms p99 latency. Key highlights: 1. Multi-DB backend support: Plug into ScyllaDB, Dragonfly, or Redis. 2. Control Plane & Admin Ops: Built-in approvals, role-based access, and audit trails. 3. Feature Cataloging UI: Manage, discover, and evolve features collaboratively. 4. Optimized for economics: Every component is designed with cost-efficiency and scale in mind.  And this is just the beginning. A whole armory of components—from stream-based feature computation to embedding search and model serving—is on its way. We're building BharatMLStack to make high-scale ML accessible, economical, and open. Dive in: https://lnkd.in/g5itp4Sk Docs: https://lnkd.in/gEimvDMx Blogs: https://lnkd.in/gZwhgrJY Quick-start: https://lnkd.in/g5zXXr27 #opensource #mlplatform #featurestore #mlops #bharatmlstack #meesho #scylladb #redis #dragonfly #machinelearning #costefficientML

385

16 Comments

I hate legacy code. You hate legacy code. Even my grandma, your grandma,  And probably the guy who wrote it hates legacy code. But in tech, legacy code is like Delhi pollution. You can complain about it all day, but at some point, you still have to breathe and get work done. After 7+ years of dealing with old functions, mystery classes, and comments that lie straight to your face, here’s what I’ve learned about growing because of legacy code. 1. Let’s not judge and criticize. Most juniors jump straight to rewriting. Seniors slow down and observe. Legacy code usually exists because it works for some use case someone once cared about. Before touching anything, read the inputs, read the outputs, check for side effects. Example: If a function is doing five random things, map them out. Often you’ll see patterns that reveal why the original engineer wrote it in that shape. This habit builds your problem-understanding skills faster than writing new code. 2. Improve behavior before improving beauty Your goal isn’t to “clean up code” but to avoid breaking the universe. Wrap the code in tests, snapshot the current behavior, then refactor. It gives you a safety net and makes you fearless. Example: I once had to touch a 900-line Python script that sent out billing emails. I didn’t touch a single line until I added a couple of input/output tests. Those tests caught three hidden issues before I even started refactoring. 3. Document what the original developers never did Legacy code forces you to become the historian the team desperately needed. Every time you understand something, write it down in simple language. This doesn’t just help others. It sharpens your own clarity and pushes you into a leadership role. Example: Create a short “What this module actually does” note. Not a full wiki, just a clear 10–15 line explanation. People will start coming to you for context. 4. Break big tangled code noodles into small, understandable units Legacy code often feels impossible because you look at it as a giant mess. Instead, Split logic into tiny blocks. Name them clearly. Move repeated parts out. Make the code readable even if it’s still old. Example: Pull one section into its own function. Just one. Next time you touch the file, pull out another. Over months, the entire module transforms. Small changes scale. 5. Treat legacy code as leadership training Legacy code teaches empathy. It teaches patience. And it teaches you how to guide others through mess. If you can explain a messy system clearly, you’re already operating at a senior level. Example: Teach a junior how a legacy module works. Walk them through it step by step. That’s how you grow from “someone who fixes code” into “someone who builds engineers.” If you can handle legacy code calmly, you can handle anything. It’s not glamorous, but it builds the skill set most engineers only learn the hard way.

513

25 Comments

If labor is cheap, does India even need robots? This is the central, uncomfortable question I explored with Pranav Srinivasan of Accio Robotics . We unpack the complex clash between India's immense human capital and its high-tech ambitions, covering everything from funding and talent to the future of manufacturing. This conversation is for anyone who believes in India's potential but wants to understand the obstacles. YouTube Link : https://lnkd.in/gUV65iwr Spotify Link : https://lnkd.in/gnnXCU7A

132

3 Comments

Twin brothers from India built an AI platform that achieved $15M in ARR in just 90 days. Is this the Shopify-for-apps moment? Here’s is the full story Emergent, founded by twin brothers Mukund (ex-Dunzo CTO) and Madhav Jha (ex-Dropbox/Amazon infra), just raised $23M Series A led by Lightspeed (total $30M to date). Their pitch? A multi-agent ‘vibe coding’ platform that turns plain English prompts into full-stack apps: frontend, backend, auth, payments, deployment, all automated. And the traction is wild: 👉 YC copy in mid-2024: 700k users, $10M ARR in 2 months. 👉 By Sept 2025: over 1M people building 2M apps and $15M ARR in 90 days. Why does it matter? Because Emergent isn’t just “no-code.” It aims to be the Shopify for apps, where anyone can generate production-ready tools, deploy them instantly, and eventually monetise them. Think jewellery chains spinning up AI-pricing apps, SMBs ditching spreadsheets for onboarding tools, even healthcare patients building pain-management solutions. But here’s the catch 👇 Retention > creation. Generating millions of apps is flashy. Keeping them live, secure, and monetised is harder. Discovery and billing flows are still in their early stages. Quality & compliance. Building real apps means adhering to PCI compliance, implementing secure authentication, and maintaining safe data practices. Emergent claims in-house testing/security agents, but independent audits will matter. Competition is fierce. Replit, Builder(dot)ai, Rocket(dot)New, Cursor: Everyone is chasing “apps from text.” Emergent’s moat must be infrastructure, reliability, and a marketplace. Unit economics. Hosting and runtime LLM costs scale with the number of active users. Using cheaper models and caching helps, but the test is focused on gross margin at scale. The opportunity: The low-code/AI app market is forecast to hit hundreds of billions by 2030+. If Emergent nails reliability, discovery, and monetisation, it could be the default platform SMBs use to generate and sell apps in days, not months. Question for you: If you could build a full-stack app from a single prompt, what would you build first? ➕ Save this post and follow Zubin Kutar ⚡ to Learn & Grow 10x with AI

189

22 Comments

Had a fascinating discussion today with a friend and fellow author who uses LLMs extensively in their work. A core question they posed was whether AI, specifically large language models, can become 'angry,' and if future AI systems might act based on emotions – a thought-provoking scenario reminiscent of science fiction. My response was based on the nature of LLMs as sophisticated pattern compressors and extrapolators of human-written language. Given that LLMs training data comprises a vast corpora of human text – full of literature, opinions, and conversations brimming with emotion – it seems inevitable that these models encode and learn these emotional patterns. While post-training alignment and techniques like Reinforcement Learning from Human Feedback (RLHF) are designed to steer model behaviour towards desired outcomes (like helpfulness and harmlessness), they primarily serve to suppress or redirect the expression of these inherent patterns, rather than eradicating them. In fact, RLHF itself optimises for human preference, which is often emotionally driven, potentially re-introducing certain emotional biases or patterns. This leads me to believe that LLMs' internal state (their weights) likely contains the capacity to generate responses tainted by the full spectrum of human emotions, including negative ones, even if alignment tries to hide or mitigate this. So, is it accurate to view LLM responses as purely mechanical, or is the presence of encoded emotional patterns from their training data an unavoidable aspect of their design? Curious to hear your thoughts. #AI #LLM #ArtificialIntelligence #NLP #MachineLearning #EthicsInAI #AIEthics #DataScience

74

11 Comments

What Aditya, Anuj and Akshay have built with Beco should ideally have gotten much more attention. And their real genius isn't just in making eco-friendly alternatives to everyday household items like trash bags, tissue rolls, and kitchen towels - it's in engineering a competitive moat so deep and integrated that legacy players and newcomers alike will find it almost impossible to cross. Having seen them aceing this game, I felt, I should write and build more awareness on this. So, to start with, the Bamboo Imperative. - Beco’s strategic choice of Moso bamboo is a masterstroke - This isn't just a greener alternative to wood pulp - it's a superior raw material Bamboo is a grass that matures in 3-5 years versus 20-50 for trees, requires less water, no pesticides, and its fibres are naturally softer and more absorbent. It's an environmental and performance win. Second, the Purity Principle. The brand promise of "NO CHEMICALS" is an incredibly powerful purchasing driver. While competitors dabble in "natural," Beco has developed proprietary technologies like "Cocozyme Technology," using plant-based ingredients like coconut and cornstarch to ensure its products are genuinely non-toxic, pet-safe, and baby-safe. This shifts the conversation from just cleaning to ensuring a healthy home environment. Third, the Packaging Promise. Their strict 100% plastic-free packaging policy is a stark, visible differentiator on every shelf, both digital and physical. In a country where almost all of us are deeply concerned about packaging waste, this isn't a feature - it's a core tenet of the brand that builds immense trust. Here’s the critical insight: a competitor might launch a bamboo tissue, but it will come in a plastic wrapper. Another might sell a natural cleaner, but it will be in a virgin plastic bottle. Beco's brilliance is the seamless, uncompromising integration of all three pillars. They have built a "system moat." Replicating it requires a complete overhaul of a company's supply chain, R&D philosophy, and core values. It’s not as easy as launching a single green SKU - it's about re-engineering the entire value chain from root to shelf. And this is what makes me see Aditya, Anuj and Akshay as such stars. And given India’s challenges, we need more and more fresh thinkers like them to turn it all around. Do try their products on quick commerce (they are growing like a rocket on this channel), on Amazon/FK or on https://www.letsbeco.com/ You’ll love them :) ---- I've begun to share my learnings from the world of startups more frequently to gain wider perspective. Thus, if you hold any additions or differences to the thoughts I shared above, do share in the comments. That will help widen my thinking horizons. Thanks! Best, Shiv

214

15 Comments

One year ago, I made the most unconventional decision of my career - I moved from India’s Silicon Valley, #NammaBengaluru, to #Dubai. It wasn’t the obvious move for someone deeply embedded in the Indian tech ecosystem. But in hindsight, it’s been one of the most expansive decisions - professionally and personally. Here are some reflections after a year of building, learning, and adapting: 🔹 What I admire more about Bengaluru (after leaving it): ✅ Tech talent density - No city in India and even Dubai come close. ✅ Business-first engineers - The depth of domain understanding and product thinking is intuitive, possibly a byproduct of the startup density. ✅ Startup ease - Access to capital, talent, and momentum is real - though differentiation and PMF remain tough challenges. ✅ The weather - You only realize its value after you leave. It’s not a perk, it’s a multiplier. ✅Hiring anxiety is overrated - The job market is more accessible than it seems — even with the rise of AI and efficiency narratives. 🔹 What Dubai gets incredibly right: 💡 The opportunity to build is massive. Sectors are open, systems are evolving, and ambition is welcome. 💡 Diversity isn’t just a buzzword. The range of people, cultures, and professional backgrounds makes you sharper, more empathetic, and more balanced. 💡 0% income tax. It genuinely changes your financial math. And your mental bandwidth. 💡 Quality of life is not a footnote. Reduced friction in everyday life makes you realize how much of mental gymnastics was going on subconsciously in your head. 💡 It doesn’t feel far from home. A 3-hour flight, massive Indian diaspora, and cultural proximity never makes you feel far from home. Grateful to the many friends and mentors who made this move smoother - shifting countries with a family is never trivial, but their guidance made it feel achievable. If you're thinking about making a similar move and want to chat (not about job referrals, but about the decision itself), feel free to DM me.

288

1 Comment