Realizing everyone viewing this post will not see the year 2100

One of the best things about remote work: freedom to travel without pressing pause on life. Monday hits differently from Fairmont Jaipur.

Online doodh ka alag hi talent hai Amazon/Flipkart/Blinkit se mangao toh garam krte hii doodh phat jata hai 🥲 Nearby shop se lo toh ekdum theek rehta hai. Anybody else facing this?

Spent the weekend building v1 of an LLM Observability Platform. When building with LLMs, provider APIs fail often. You have to decouple the fast path (streaming inference) from the slow path (telemetry). This architecture uses Spring WebFlux for non-blocking SSE streams, Kafka to offload lifecycle events, and ClickHouse for OLAP analytics. It's fully containerized for local testing. I’m currently iterating on the design—would love suggestions or feedback on the architecture. github.com/pankaj45/llm-obse… #SystemDesign #SoftwareEngineering #Java #LLMOps

One nation, one flirt policy.

Bro really said: “Roses are red, violets are blue… Melody khaogi?” 🍬

Learn from Claude. Nobody hypes you up at work? Do it yourself. “Excellent research. Now I’ll write the document.” That AI has better self-confidence than 90% of software engineers.

Infinite money glitch 😂 PureVPN is currently giving 120% cashback via TopCashback. Meaning: Buy PureVPN Get more cashback than what you paid You literally get paid to use a VPN These cashback stacking opportunities are why I always check TopCashback before buying anything online. Here's the screenshot from my yesterday's purchase. Bought the 2 year plan. P.S: Cashback will get confirmed within 3 months. That's the normal timeline. topcashback.com/purevpn/

If you’re building something ambitious and looking for a technical co-founder or founding engineer, DM me. 9 years building backend/distributed systems with Java, Golang,Python, AWS, Kubernetes, databases, infra, and now deeply focused on AI/agentic systems. I enjoy going from 0 → 1: product architecture scalable backend systems AI integrations/agents developer infra shipping fast without creating a mess later Especially interested in AI, developer tools, infra, fintech, and workflow automation.

My job as a staff engineer has quietly evolved into: translating vague product thoughts into prompts Claude won’t creatively reinterpret. reviewing architecture proposed by agents that are extremely confident. letting Jira MCP generate 15 tickets I’ll later merge back into 6 actual workstreams. acting as the human “are we seriously shipping this?” checkpoint for AI-generated MRs. debugging why 4 coding agents implemented 5 different understandings of the same requirement I definitely write less code now. Most of my job is maintaining enough context, constraints, and guardrails so autonomous code generation doesn’t turn the codebase into a distributed systems thought experiment.

iPhone after 80% battery: “AC on karo, cover hatao, mujhe hawa do… tab charge hoga.” Bhai phone charge kar raha hu ya honeymoon package de raha hu? 😭📱

Ordered soya on Swiggy. Restaurant sent chicken instead. I’m vegetarian by choice, not because of religious beliefs, which is why I’m trying to stay calm about it. But I had already taken the first bite before realizing it. Swiggy refunded the amount on escalation, but I’m genuinely wondering if I should file a formal complaint so this doesn’t happen again with anybody. Veg/non-veg is not a small mixup. For many people it’s religious, cultural, and deeply personal. One careless restaurant zero verification from delivery platforms can seriously hurt sentiments.

Most people add indexes. Few design them. A covering index is when your query can be answered entirely from the index → no table lookup. Think of it like: You asked for a book → librarian hands you summary card → you don’t need to fetch the whole book. Example SELECT name, age FROM users WHERE city = 'Delhi'; Index: CREATE INDEX idx_users_city_name_age ON users(city) INCLUDE (name, age); Now DB never touches the main table. Faster. Fewer IOs. Happier CPU. Senior engineers know: 👉 This is where real latency wins come from 👉 Especially in high-read systems But tradeoff? More storage slower writes. Where have you used covering indexes in production?

I interviewed with a German startup for a Staff Software Engineer role. 4 rounds. 8 hours of interviews. Every round: positive feedback. 10 days later (Today) → rejection. Reason: “Other candidate had more AI experience.” Interesting how when candidates keep multiple options, it’s called offer shopping. But companies doing the exact same thing is framed as “hiring the best fit.” End of the day, both sides are optimizing. Only difference is one side gets judged for it more.

It’s been more than 30 hours. Still internet is not fixed. @excitel_rocks Shame on you. Customer care people also have no clue.

I told my friend today, “Bro, quit smoking… your lungs are gonna get wrecked.” He goes, “And what’s the benefit of dying with perfectly healthy lungs?” 😂😂💀

Most system design talks sound smart… until you ask for numbers. Here’s a practical cheat sheet engineers actually use: Cache ~1ms latency, 100k ops/sec ⚠️ Scale when: hit rate <80%, latency >1ms, memory >80% Database ~50k TPS, <5ms reads (cached) ⚠️ Scale when: writes >10k TPS, uncached reads >5ms, geo needs grow App servers 100k connections, CPU sweet spot <70% ⚠️ Scale when: latency breaches SLA, connections spike, memory >80% Message queues ~1M msgs/sec, <5ms latency ⚠️ Scale when: throughput ~800k/sec, partitions explode, lag builds up These aren’t hard limits. They’re signals your system is about to hurt. Good engineers monitor metrics. Great engineers anticipate the breaking point.

Ever seen your system melt because 10k requests asked for the same data at the same time? Classic mistake: all 10k hit DB/cache independently. Senior move: Request Coalescing - first request does the work, rest just wait and reuse the result. Like one person ordering for the whole table instead of 10 people shouting at the waiter. Here’s a clean way to do it in Spring Boot 👇

You’re processing a payment with a Redis lock. Your service pauses, lock expires, another worker retries the same payment. 👉 How do you ensure the first (stale) worker doesn’t still update the DB and double-charge the user?

Tinder in Bengaluru 🤓

Choosing the right database can make or break your project. Here’s a simple way to think about it 👇 Start with one question: What is your primary workload? 1️⃣ Analytics & Big Data If you’re running heavy analytics, dashboards, or BI queries, use a Data Warehouse. Good options: Snowflake or BigQuery. 2️⃣ Operational / Application Data If you’re building a typical app (users, orders, products), ask: • Need ACID transactions? • Have complex joins & relations? If yes → use a Relational DB like PostgreSQL or MySQL. They’re reliable, structured, and great for most applications. If you need high write scale distributed SQL, consider NewSQL like CockroachDB. 3️⃣ Schema Flexibility? If your data structure changes often: • Document DB → MongoDB (great for JSON-style data) • Graph DB → Neo4j (perfect for relationships like social networks) 4️⃣ Access Pattern Matters • Wide column workloads / massive scale → Cassandra or DynamoDB • Ultra-low latency caching → Redis or Memcached 5️⃣ Specialized Use Cases Some problems need purpose-built databases: • Search → Elasticsearch • Time-series data (metrics, IoT) → InfluxDB • Vector search (AI / embeddings) → Pinecone or Milvus • File/object storage → S3 or GCS Start simple. In most cases, PostgreSQL is enough for your first version. Optimize later when your workload demands it. The best database isn’t the most popular one, it’s the one that matches your data model, scale, and access pattern. What database do you usually start with for new projects?