Writing...
Thoughts on software engineering, tech stacks, and building products.
The AI Boom Is a Backend Problem
The world will spend about $2.7 trillion on AI this year, and only about two in ten companies say they're scaling agents. I read this autumn's market reports as a backend developer, and most of the gap between those two numbers is retries, token budgets, evals and permissions.
The Quote Was ₹55. The Bill Was ₹175.
A shopper paid ₹55 for delivery and the courier billed ₹175, on every order. Nothing errored. The parcel was just described two different ways. A deep dive into volumetric weight, rounding, cache keys, and keeping a delivery quote honest.
Working, I Guess
Three words in my bio, one Sunday to split between them. A short note on what indie hacking looks like when you are also a student: less building, more hoping the email actually sent.
The Wrong Answer Is the Interesting Part
Most quizzes grade you and move on. Kuiz Space is built on the opposite idea: a wrong answer tells you more than a score, so every option explains itself.
Progress Is Not a Field
I built ProgresTracker around one slightly annoying rule: nobody gets to type a project's completion percentage. It has to be earned by the tasks underneath it.
Designing a Rate Limiter (and the Ways I Got It Wrong First)
A request-per-minute cap sounds like a one-line if statement. It isn't. A deep dive into fixed windows, sliding windows, token buckets, and why the real fight is over where the counter lives.
The Bug That Made Resumter Actually Useful
I built a resume analyser, fed it my own resume, and watched it complain about a missing GitHub link that was sitting right there on the page. The fix taught me more than the feature did.
The Hackathon Where I Built a Router Instead of a Tutor
Everyone at the hackathon was building an AI tutor. I built the thing that sits between the tutor and its tools. Here's what that looked like, what LangGraph gave me, and what I'd do differently.
I Tried My Own Context Engineering Advice. It Went Badly.
Last week I wrote five neat rules for managing an LLM's context window. Then I applied them to a real agent and broke three things. Here's what the rules left out.
Context Engineering: Your Prompt Isn't the Problem
Most bad LLM output isn't a model failure, it's a context failure. A practical guide to budgeting, retrieving, and compacting what you actually send to the model.
The Prerequisite Trap
The coffee was hot, the terminal was clear, and the goal was set: today was the day I finally conquered System Design and Backend Systems.
A Simple Tool Calling
AI applications is not just wrapper, its is agentic systems, and tool orchestration is the key to unlock the full potential of AI agents.
Building Book Your Show: A Robust Cinema Booking Backend
A deep dive into the architecture and features of the Book Your Show backend, built with Spring Boot and PostgreSQL.
How to Learn Spring Boot in 2026
A strategic guide to mastering Spring Boot 4, Java 25, and cloud-native development in 2026.