All projects
TypeScript AI API
Production-ready Fastify REST API in TypeScript, integrated with the OpenAI API to deliver intelligent-assistant capabilities behind clean, typed endpoints.
Fastify, TypeScript, OpenAI API, Node.js, Zod, Vitest
The problem
Teams integrating an LLM into a product need a small, fast, well-tested API layer in front of OpenAI — with auth, rate limiting, and validation already handled — rather than wiring raw SDK calls into every service.
My role
Responsibilities
Built the API solo: route design, the AiService wrapper around OpenAI, the auth middleware, and the full Vitest test suite.
Main workflows
- 1.Client authenticates every AI request with an X-API-Key header
- 2.POST /ai/chat proxies a multi-turn conversation to an OpenAI GPT model and returns the assistant's reply plus token usage
- 3.POST /ai/summarize condenses long text into a paragraph or bullet-point summary
- 4.GET /health reports API and AI service status for uptime checks
Architecture
- Client app
- Fastify REST API (TypeScript)
- OpenAI API
- Fastify chosen over Express for 2–3x throughput, native TypeScript support, and schema-based validation
- Zod schemas validate every request body at runtime, mirroring the TypeScript types with clearer error messages than raw JSON Schema
- Service layer pattern: AiService is injected into routes, so it's mockable in tests without real HTTP calls to OpenAI
- App factory pattern (buildApp()) returns a fresh Fastify instance per test, avoiding port conflicts in the suite
Security and deployment
- X-API-Key header required on every AI route
- 60 requests/min per IP via @fastify/rate-limit
- Security headers via @fastify/helmet
- Multi-stage Dockerfile + docker-compose for containerized deployment; auto-generated Swagger docs at /docs
Results
Full unit + integration test suite via Vitest, covering routes, the service layer, and error handling for 400/401/403/404/500.