Puneet Patwari

I interviewed at Google, Uber, Walmart, Amazon and several top startups (a total of 60+ interviews) during my job search from March to June 2025 before joining Atlassian as a Principal Engineer. Here’s what each experience taught me. If you’re prepping for a switch, I hope this gives you clarity on what to expect (and how to survive the roller coaster):

@DesignCodeCo @Star_Knight12 Sarcasm 🙄😂

@system_monarch @Star_Knight12 Happy ending? Are u ok?

once AI will be able to cure cancer, all AI haters will go silent

@SumitM_X "Acha chalta hu, Duaoo me yad rakhna"

What will be your PR review comment if you see this line in the PR : Thread.currentThread().join();

@arpit_bhayani Thanks for the clarity 😌

If you are deduplicating documents, semantic similarity is probably the wrong tool for the job. I am building a tool to manage my second brain, and there I need to de-duplicate documents (thoughts). Instead of using vector similarity, I am using Jaccard similarity. Here's why... Semantic similarity (via embeddings) captures meaning. Two documents can score high similarity even when they share almost no words, because they are talking about the same topic. That is great for search and recommendation, but it is terrible for deduplication. Jaccard similarity measures the overlap of words or tokens or shingles (short character sequences) between two documents: how many they share versus how many are unique to each. It does not care about meaning at all. It cares about surface-level overlap. That is exactly what you want when finding near-identical documents. A scraped article and its slightly edited copy will share 80-90% of their 5-grams. Jaccard catches that. An embedding model might not even flag it. The right mental model: use semantic similarity to find documents that are about the same thing. Use Jaccard to find documents that are the same thing. Hope this helps.

@SumitM_X Database = OLTP Data warehouse = OLAP Data lake = Data Dumpyard

As a developer, Do you know what is the difference between : Database, Data Warehouse and Data Lake ?

@SumitM_X Started with TCS... 🙂

There are 4 types of software engineers: - Those who start with Service based MNCs like TCS, TechM, CTS, Accenture, and Wipro. - Those who start with Product based MNCs. - Those who start with captives - Those who start with start-up Which one are you ?

@elonmusk No AI can create such a beautiful scenery 🥹

@knowRowan World was about to end in COVID but out came AI revolution out of nowhere. So we should have hope 😅

@system_monarch Like?

Hot take: Ai will lower the IQ of the human race

@DMSCoding11 Job at Google. I feel it's still the pioneer.

Which one would you choose: 1. Job at Google 2. Job at Meta 3. Job at Apple

@RoundtableSpace Building my website to offer my services at puneetpatwari.in

What are you building today?

@elonmusk Video so good and clear, it looks like AI.. 😜

@yashhq_22 The life hack no one shared. Thanks for the insight. 😅

Posting 3x a day on X won't grow your account. Replying 100x a day will. Nobody wants to hear it though.

@TTrimoreau Improve onboarding & talking to churned users will reveal surprising stuff about the gaps.

You just launched your SaaS. Users sign up… but no one comes back. You can only choose one move: -Improve onboarding -Add features -Send emails -Talk to churned users What are you doing? 👇

@dharmvir_ Hand's down Youtube. The best generalist and mostly free.

Which is the Best platform to actually learn and build?

@sickdotdev Mentoring, Software Engineering.

Name a skill that actually makes you money

@TTrimoreau Did from GoDaddy but so many ADs and quite slow UI.

Founders where do you usually buy your domains -GoDaddy -Dynadot -Hostinger -OVHCloud -Ionos And why ? 👇 because need a new one

@lunarfq Hello 👋

A lot of people are struggling to get 500 verified followers for monetization while I have about 6.4k verified mutuals out of 31.5k Do yourself a favour and say Hello, let them follow NOW.

Backpressure stops your service from drowning. But how do you prevent clients from flooding you in the first place? That’s where "distributed rate limiting" comes in and most candidates treat it like a simple local counter. Here’s the Principal-level way I think about it: 1. Choose the right algorithm → Token Bucket (flexible bursts) vs Leaky Bucket (smooth flow) vs Fixed Window (simple but bursty) 2. Make it distributed → Never use in-memory counters alone. Use Redis (or similar) as a shared counter with Lua scripts for atomicity. 3. Handle hot spots & consistency → Shard by userID/clientID. Accept eventual consistency for most cases, but strong consistency for critical paths (e.g., login abuse). 4. Layer it properly → Edge (CDN/API Gateway) for coarse limits + Service layer for fine-grained per-tenant or per-feature limits. 5. Add observability & graceful handling → Track rejection rate, latency impact, and allowed vs throttled requests. Return 429 with Retry-After header + exponential backoff guidance. At Atlassian we combine rate limiting + backpressure + circuit breakers to keep Jira/Confluence stable even during massive spikes. The full deep dive on rate limiting algorithms, sharding trade-offs, and failure modes is inside my System Design Fundamentals Guide. → Free sample + 90+ fundamentals here: puneetpatwari.in What’s the trickiest part of rate limiting you’ve faced in production or interviews? Reply below 👇 #SystemDesign #DistributedSystems #StaffEngineer #RateLimiting

@TTrimoreau Being relentless..

As a founder What is the one thing that separates winners from the rest?