⚀ I build backends that don't fall over, and frontends like this one.
⚁ Rust, Go, and the boring reliability work most people skip.
⚂ AI tools, shipped and in production. Not demos.
I love crafting production-grade software that solves real problems.
Demo
Turns raw recordings into speech-model training data, with a reason for every clip it drops. Caught 24,385 of 24,409 injected defects across six languages, curates audio 940x faster than real time on a laptop CPU, and judges 1,000 live streams at 14.8 ms p95 over WebSocket. Runs in the browser via WebAssembly, and ships as a CLI, Rust crate and Python bindings.
Tech Stack
Experiences
FinStocks AI
Lead Software Engineer
- I was the first hire. Now I lead a team of 3, and we built the thing that turns "buy when RSI dips under 30" into a strategy you can actually backtest.
- The LLM never touches the math. It just fills out a typed spec (185 indicators, 148 patterns) and a compiler takes it from there. Same spec, same result, every time.
- Backtests chew through the whole stock universe in one pass with Numba-JIT kernels, and only read the Parquet columns a strategy actually needs.
- Backtesting and live trading share one compiler, so what you tested is exactly what trades. Redis Lua scripts stop workers from fighting over capital.
- Built a chart terminal you can just talk to. 85 typed operations, plus a 1.4k-line validator whose whole job is stopping the LLM from inventing price levels.
- Push notifications run on Postgres with SKIP LOCKED, so nothing gets lost and nobody gets pinged twice.
- Market-data queries went from 13.6s to 48-135ms, and another one from 5.9s to 782ms. Live bars stream over Redis Pub/Sub straight into TradingView charts.
Indian Kanoon
Software Developer
- Built Prism from scratch. It's AI that reads case law and legal documents at scale.
- Got it handling 5,000+ concurrent users.
Chargebee
Software Engineer Intern
- Led a big data migration and a round of database tuning.
- Moved 2 million records and cut query time by 45%.
- Built a validation layer that caught bad data before it reached 500k subscriptions.
AiDash
Software Engineer Intern
- Designed APIs and data retrieval frameworks built to scale.
- Broke a monolith into microservices.
- Tuned SQL pagination until retrieval felt instant.
Proof Of Work
building real tools, solving real problems, and leaving a trail of commits to prove it.
Contributed to