ECE @ NSUT Delhi '29 · I build with AI agents in the loop — and make them real: full-stack, deployed, with the boring parts (auth, databases, deploys) actually working.
Smart India Hackathon 2026. I owned the crypto verification core (JWS-signed QR codes) and the PDF generation pipeline. Six-person team — live in production, not a prototype.
| Hackathons | NHAI Hackathon 7.0 finalist (CV liveness module) · Hack4Delhi (RAG) · Hack2skill · HackAgentAIx — I sign up for too many of these |
| Agent tooling | I write skills for coding agents — youtube-playlist-skill turns any topic into a real public YouTube playlist, installable straight into your agent |
context.txtfrozen contract → per-person prompt books → human-reviewed merges
Our SIH sprint ran on a workflow I set up: a contract file everyone reads before touching code, prompt books so each member's agent knew exactly what it owned, and humans reviewing every merge. Shipped in days what normally takes weeks. I like building the workflow as much as the product.
| Focus | What it looks like |
|---|---|
| DSA grind | C++, every day. First 20 minutes of every problem are mine, no agent help — the OA room won't have an AI copilot in it |
| PyTorch, properly | "I've used LLM APIs" and "I know ML" are different sentences. I'm earning the second one |
| PRAMANAM | Keeping it alive after the hackathon — uptime, fixes, real users |
| Level | Tools |
|---|---|
| Daily | TypeScript · Next.js · Python · PostgreSQL · Git |
| Dangerous | FastAPI · Redis · Docker · C++ |
| Learning for real | PyTorch · ML fundamentals |