Kunal Roy Choudhury

⚀ 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.

Gostman Demo

Open source API client for REST, GraphQL, gRPC and WebSocket: a native desktop app (Go + Wails), a web app, and a single-binary CLI that runs collections in CI and serves them to coding agents over MCP. No account, no telemetry, no paid tier.

Tech Stack

AeroSieve

Turns raw recordings into speech-model training data. Decodes audio, cuts it into speech clips, drops the silent, noisy, clipped, or tonal ones, and writes WebDataset shards. 43 minutes of Hindi audio curated in 1.7s on one CPU core. Ships as a CLI, Rust crate, and Python bindings.

Tech Stack

Show 5 more projects

Experiences

  • First employee here. I lead a team of 3 and built the engine that turns plain English into trading strategies you can backtest.
  • The LLM only fills in a typed Pydantic spec covering 185 indicators and 148 patterns. Everything after that is compiled and validated without it, so results are deterministic.
  • Made backtests fast with Numba-JIT kernels that run the whole universe in one pass over float32 Parquet, loading only the columns a strategy actually uses.
  • Backtests and live trading run on the same compiler, so what you tested is what trades. Redis Lua keeps capital allocation atomic across workers.
  • Built an LLM chart terminal with 85 typed operations. A 1.4k-line validator stops it from making up price levels.
  • Built transactional push on Postgres with SKIP LOCKED, so notifications don't get lost or sent twice.
  • Took market-data queries from 13.6s down to 48-135ms, and from 5.9s to 782ms. Bars stream over Redis Pub/Sub into a TradingView datafeed.
  • Built Prism from scratch. It's AI that reads case law and legal documents at scale.
  • Got it handling 5,000+ concurrent users.
  • 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.
  • Designed APIs and data retrieval frameworks built to scale.
  • Broke a monolith into microservices.
  • Tuned SQL pagination until retrieval felt instant.

Get in touch

Hi there, I'm currently open to meaningful work.