2025 · Solo — prompt design, backend, deployment
Code Review Assistant
An LLM agent that reviews pull requests and returns structured, actionable feedback.
- Output
- Structured JSON
- Status
- Deployed live
- Model
- Google Gemini
the problem
Review bots get muted within a week. The hard constraint is not finding issues — a language model will find you an unlimited supply — it is returning few enough, and specific enough, that people keep reading them.
the approach
The agent analyses pull requests and code snippets through the Google Gemini API, with prompts constrained to emit structured JSON rather than prose, so every finding carries a location and a category that the client can filter on. The backend is Node.js, Express, and TypeScript, built along clean REST and service-layer lines, with request rate-limiting at the edge and Supabase for persistence.
the result
A live, publicly usable tool that I own end to end — prompt design, API integration, backend architecture, and deployment. Structured output is what makes it useful: findings can be ranked and suppressed rather than dumped into a thread.
stack
- TypeScript
- Node.js
- Express
- Google Gemini API
- Supabase
- REST