Founding engineer at olum.ai, where I own the frontend and the backend APIs. Moving toward backend and distributed systems - currently writing Go and reading DDIA.

LinkedIn · LeetCode · barathraj048@gmail.com

Founding engineer, olum.ai. I own the product surface end to end - the frontend and the backend APIs behind it. Small team, so the API contract, the schema, and what the user sees are all mine to get right.

• Olum is an AI-SEO platform that audits a site, then continuously tracks and improves how it surfaces in both Google and AI assistants like ChatGPT - generating the content and fixes to close the gaps.
• ~70 endpoints across 12 service groups (auth/billing, analysis pipeline, competitor and social intelligence, AI-visibility tracking) behind a 60-page React app, ~79k LOC; long-running crawl/LLM work runs async with client polling, so interactive reads stay under a ~500ms p95 while analysis runs finish in minutes.
• Argued for DPoP (RFC 9449) sender-constrained tokens on the payment path instead of plain bearer auth: a non-extractable P-256 keypair in IndexedDB signs a fresh proof per request, and the backend matches its thumbprint against cnf.jkt on the token. Our access token already rides in an httpOnly cookie, so XSS can't read it - but it can still call the API from the victim's page; binding the token to a key JS can never export makes a stolen token useless elsewhere. Cost was a refresh-flow rewrite and a WebCrypto dependency, so I scoped it to payments rather than every route.

Only put a number here if you measured it. One real number beats three vague ones.

Go · DDIA (2nd edition) · storage engines · consensus.

I build APIs at work, so the depth is coming from projects: a key-value store in Go with an append-only log, CRC-checked records, an index rebuilt from disk on startup, and crash recovery that survives kill -9 mid-write. Benchmarks will include the numbers that aren't flattering.

1,797 contest rating - top 8.4%, ranked 72,294 of 879,441. 52 contests attended.

329 problems solved: 207 medium, 26 hard. 156 active days in the past year.

n8n (230k+ active global users) - five pull requests, none merged. Four were triaged as valid and assigned to internal teams (#28561, #29203, #29415, #30151). What I found: a Date serialisation bug in core deepCopy, a gap in hybrid dot/bracket notation in the secrets parser, cross-platform path handling that broke Windows CI, and workflow layout coordinates being dropped on save.

Finding real bugs in a 200k-star codebase was the easy half. Landing a patch in a subsystem with an internal owner is the hard half, and I got that wrong five times before I understood why - untested diffs, scope too wide, and writing code before a maintainer had replied. I now comment on the issue and wait for agreement before opening anything.

Cal.com (1M+ signed-in users) - one merged pull request (#27251): a padding fix on VerticalTabItem.

In-memory trading engine - Node.js, Redis, WebSockets. Order matching with an in-memory book. Load-tested at 1,000 req/s across 500 virtual users: p95 21.9 ms, 0 errors.

Distributed task pipeline - Redis producer/consumer with per-task idempotency. Pub/Sub and WebSocket fan-out instead of HTTP polling. Containerised with Docker.

Care Ops - healthcare ERP, built in a 48-hour hackathon (top 10). 367 req/s at 100 virtual users, no timeouts under load.

Languages TypeScript · JavaScript · Python · Go Frontend Next.js · React · Tailwind Backend Node.js · Express · WebSockets · Pub/Sub · REST API design Data PostgreSQL · Redis · MongoDB Infra Docker · Linux · Vercel

B.E. Electronics & Communication Engineering, P. A. College of Engineering and Technology (Anna University), 2022–2026.

Student Chairperson, IETE - 18-member team.