Work

Projects

Systems built from scratch — rate limiters, code runners, and AI pipelines. Each one solving a specific problem with deliberate trade-offs.

Throttle

Standalone rate-limiting service. Supports Token Bucket, Sliding Window, and Fixed Window algorithms — each backed by Redis Lua scripts (EVALSHA) for atomicity across concurrent instances. Any application in any language can delegate rate limiting via a single HTTP call, keeping policy logic out of individual services.

Go · Redis · Lua · PostgreSQL · Docker

GitHub

CodeBench

Remote code execution platform. Python submissions run in isolated Docker containers with hard resource caps — no network, 128 MB RAM, 0.5 CPU, 10s timeout, read-only filesystem. A 202 + polling architecture via Sidekiq keeps the web process non-blocking. On acceptance, Gemini generates three follow-up questions specific to the submitted code.

Ruby on Rails · React · Sidekiq · Redis · PostgreSQL · Docker · LLMs

GitHub

Haqq

AI-powered legal assistant. Users describe situations and receive plain-language legal analysis and document drafting against an ingested corpus stored in pgVector. FastAPI handles the REST layer; Redis RQ runs retrieval and LLM inference asynchronously. Five endpoints — analyze, draft, debug, examples, feedback — with per-IP rate limiting and Docker Compose orchestration.

React · Python · FastAPI · PostgreSQL · pgVector · Redis · TypeScript · Docker · LLMs

GitHub