Python/FastAPI Backend Engineer for AI-Powered Product (Frontend Already Built)

We're a 4-person team building a research assistant tool for legal and compliance teams. Our frontend is done and looks great - Next.js 14 (App Router), TypeScript, Tailwind, deployed on Vercel. Both of our engineers are frontend-focused. We've been faking the backend with mock data and a couple of thin API routes, and we've hit the wall. We need someone to own the backend properly. What we need built: - FastAPI service to replace our current Next.js API routes - Streaming endpoints for our chat interface (SSE) - right now responses appear all at once after 20+ seconds and it feels broken - Auth: currently using Clerk on the frontend, need proper token verification and per-org data isolation on the backend - Background job processing - users upload documents (PDF/DOCX, sometimes 200+ pages) that need parsing and embedding without blocking the request. We assume Celery + Redis but we're open - Postgres schema design and migrations. We have a rough schema that we know isn't right - Vector search over uploaded documents. We're on Supabase/pgvector currently, happy to move if there's a good reason - Deployment setup (we're thinking Railway or Fly.io) and basic monitoring so we know when something breaks About the AI side: We're calling the OpenAI API directly and it mostly works, but retrieval quality is inconsistent and we don't have a good way to tell whether a change made things better or worse. If you have opinions here we'd like to hear them, but the backend infrastructure is the priority. What we're looking for: - Strong Python and FastAPI, with production systems you can point to - Comfortable with async Python - a lot of this is I/O bound - Experience with background job queues at real volume - You've worked with a frontend team before and can define a clean API contract instead of us guessing - Enough familiarity with Next.js/TypeScript to read our code and tell us when the problem is on our side - Clear communicator. We're async across three time zones Nice to have: - RAG/vector search experience - Experience with Clerk or similar auth providers - You've dealt with LLM streaming and cancellation before How we work: Async-first, Slack. We'll give you full repo access and a Loom walkthrough of the product on day one. To apply: Tell us about a backend you built that a frontend team consumed - what the API contract looked like and what you'd do differently now. Please skip the generic proposals, we'll know. Screening questions: 1. Describe your experience with FastAPI and async Python. 2. Have you implemented SSE or WebSocket streaming in production? What went wrong the first time? 3. What's your approach to background job processing for long-running tasks? 4. What is your availability over the next 3 months?