Building Scalable Backend Systems

My primary stack is Nuxt 3 / Nitro with TypeScript, PostgreSQL via Prisma, Redis for caching and sessions, and BullMQ for async job queues. I also work with FastAPI (Python) for AI service endpoints.

I use domain-driven service boundaries, containerization, and async job queues to keep the request path non-blocking. Stateless APIs with Redis-backed sessions scale horizontally without sticky sessions.

PostgreSQL with Prisma for typed, migrated schemas. Redis for caching, rate limiting, and pub/sub. Weaviate for vector search. No cross-domain direct table access — each domain owns its own data layer.

JWT access tokens with short TTLs and Redis-backed refresh token rotation. Server-side session revocation via Redis. Rate limiting applied at the middleware layer before any handler executes.

Yes. WebSockets for message sync and event broadcasting. WebRTC for peer video via a Nitro signaling server. Redis pub/sub for cross-instance event routing so state stays consistent regardless of which node a client hits.

Uploads go to MinIO via pre-signed URLs. BullMQ dispatches FFmpeg transcoding jobs. Processed output is written back to MinIO; the client is notified via WebSocket on completion.

Strict TypeScript across frontend, Nitro server, and shared utilities. Zod schemas at API boundaries generate both runtime validation and inferred TypeScript types. Prisma types flow from the schema migrations.

Yes. Backend services dispatch to Llama (on-cluster) or OpenAI via a custom orchestration layer, with BullMQ queuing AI batch jobs. Results stream back to the client via SSE.