I build things people use at work.
Full-stack engineer, 4 years in. I work on the whole thing: the screen you see and the server behind it.
I have been at Tracxn since 2022, now as a Senior Software Engineer, out of Bengaluru. Most of what I do these days is AI: keeping Tracxn’s assistant accurate, fast and cheap, and the Model Context Protocol server I built for coding agents.
Happy to talk about AI, LLMs, MCP, agent tooling, architecture or system design.
Career
- Analysing Tracxn's AI Assistant logs daily for output quality, usage and cost, tracking the input and output tokens behind every turn, and optimising prompts, context and token use to cut cost and response times.
- Adding skills and new capabilities from across the Tracxn platform for the assistant to use, and checking tool behaviour to keep output quality high.
- Running POCs for new assistant features and new models, with evals comparing quality, latency and cost against the current setup, and taking feature, model and router-model changes from proposal through review to rollout.
- Built an internal MCP server from scratch, wiring Tracxn workflows into VS Code, Cursor, Claude, Codex, and ChatGPT for AI-assisted scripting, legacy API migration, and conversational access to API schemas, filters, sorting options, and response structures.
Work
A map of the startups and product companies hiring in Bengaluru, with a pin on every office. Each role is read off the company's own job board, and apply sends you straight back to it. A map, not a middleman.
hiringmap.inNext.js, Cloudflare Workers, D1, Drizzle, MapLibre
The assistant is Tracxn's, built by a team. Each day starts in its logs: how good the answers were, how much people used it, and what it cost, down to the tokens in and out. I add skills and new capabilities from across the platform, and trim the tokens and time a turn takes. I build proofs of concept for new features, and run evals and cost comparisons on new models before the model or its router changes.
TypeScript, Node.js, Vercel AI SDK
A CMS I built from scratch inside Tracxn's docs site. Authors write next to a live preview and publish when they are ready. The next deploy turns each published entry into a real page, with its own sidebar row and search entry. The docs stopped waiting on engineers.
TypeScript, Next.js, MDX, MongoDB
A read-only Model Context Protocol server I built from scratch. It gives a coding agent the shape of Tracxn's APIs: endpoints, schemas, filters and sorts. Scripts and migrations get written against the real thing, and the server never runs a live call itself.
TypeScript, Node.js, MCP, OAuth
A reading companion I built and run. You save the words and quotes you meet in a book, and it brings each word back on a schedule until it sticks. The AI runs on your own key.
bookbrain.co.inTypeScript, Next.js, Prisma, PostgreSQL, Vercel AI SDK