Build a FastAPI Backend with AI API Integration

Description

We have a React front end and, at the moment, a backend that is three Flask files one of our founders wrote in a hurry. It works for the 200 users we have. It will not work for the 2,000 we expect after launch, and nobody wants to touch it. We want it rebuilt in FastAPI, keeping the same API surface so the front end does not change: auth, the document endpoints, and the LLM calls that currently block the request thread and time out under any real load. Those need to be async with proper job handling. Provider-agnostic, please. We use OpenAI today and we have been burned by rate limits twice, so we want the ability to fail over without a rewrite.

• Rebuild the API in FastAPI against the existing contract so the React client is unchanged
• Move LLM calls off the request thread into async jobs with status polling
• Abstract the provider so a second one can be added or failed over to without touching call sites
• Migrate auth and sessions with no forced logout for existing users
• Load test to 2,000 concurrent users and report where it breaks

• FastAPI backend deployed, serving the existing API contract
• Async job handling for LLM calls with client-visible status
• Provider abstraction with a documented failover path
• Load test report and a scaling note