Headline

Language & Agents @ DeepSource

About

An engineer working on applied AI and ML; the messy part where models have to actually work in production, not just in a notebook. I care less about the demo and more about whether the thing holds up when people start relying on it. Outside work I'm usually tinkering with how models get built and optimized. Reading, implementing things, figuring out why they work. Always happy to connect with people building in the same space.

Location

Bengaluru

AI stack

antigravity

codex

Work experience

Software Engineer at DeepSource, Jan 2026 - Present · 8 mos, Bengaluru, Karnataka, India, - Built our secrets-detection pipeline end to end: it scans code for credentials, classifies what's real, and validates the hits live. I wrote my own benchmark to test it and got it to around 99% precision, since in security a false alarm wastes people's time almost as much as a miss costs them. - Monitoring under Load: an LLM endpoint kept buckling under load. Trimming what we actually sent it dropped the failures by orders of magnitude. Not glamorous, but it's the kind of fix that keeps things quietly working. - Wrote our OpenSSF CVE Benchmark submission and the evaluation harness behind it. We hit zero false positives across 165 real CVEs, the only tool to do so, against Claude Code, Cursor, Codex, CodeRabbit, and others. - Work on the multi-agent system behind our code review. An orchestrator runs specialized agents in parallel, everything's typed front to back, and a registry lets us swap models or providers in a single line.

Software Engineer at DeepSource, Mar 2025 - Dec 2025 · 10 mos, Remote

Education

Bachelor of Technology - BTech, Computer Science, Shiv Nadar University

Delhi Public School - India

Public links

github: https://github.com/unnxt30