I’m Nithin.
Former eSports player turned software engineer who works the whole stack, from pixels to distributed systems.
I’m interested in.
- Voice first interface.
- Graph-based retrieval systems for deterministic search.
- Language servers and ASTs.
- Distributed systems.
Founder journey.
In early 2026 I co-founded Basegraph. We built Relay, an AI teammate for engineering teams that works across Slack, Linear, and GitHub. It reads the team’s graph of who owns and decides what, routes one request to the right people, and DMs each in their own thread with role-fit context.
It was a fun journey. We sunset it after a few months. Adoption stalled on trust, since we were an unknown team. And the model providers moved into agent orchestration themselves, which commoditized an independent layer like ours.
My past work.
- Led Voicebot team of 4 engineers at Convin, a Series A startup. Re-architected and scaled the Voice AI platform 33x (30K to 1M calls/day, ~24M/mo) with zero-downtime releases.
- Diagnosed and fixed a critical reliability bug in Convin’s call pipeline affecting ~25% of daily calls across all clients, recovered them, and drove a major CSAT improvement org-wide.
- Led Account Aggregator Integration at Perfios, a $1bn fintech unicorn, converting one-off decision lending reports into recurring ones for financial institutions in APAC regions. Resulted in 300% increase in reports volume.
Problems I’ve solved.
- relay: an AI teammate that owned engineering work end-to-end across Slack and Linear. It read the code, found who owned what, and pulled the right people in itself. Built on a tiny kernel and a graph of markdown files. How it worked, and why we sunset it.
- codegraph: built one of the first code context graphs for Go in 2023, because I felt vector search wasn’t right for code search. It reads the compiler’s type checker so a coding agent queries exact relationships instead of guessing from text. 13x faster and 14x cheaper than the top agents at blast-radius analysis.
- dtpx: a distributed downloader that splits one file across several machines and pulls it in parallel. The idea, how it worked, and the parts I got stuck on. (paper)