Media planning and brand matching, with AI doing the analysis rather than a spreadsheet.
Backend, LLM layer, and frontend, shipped as one product. An async FastAPI service handles request orchestration, with Celery and Redis carrying the long-running analysis work off the request path so the UI stays responsive while jobs run.
The model layer routes across OpenAI and Gemini with automatic fallback on error or timeout. For a live product, provider availability is an operational concern rather than a design-time choice, so failover is built into the abstraction rather than handled per call site.
Analysis streams to the React SPA over SSE, so users see partial results as they are produced instead of waiting on a spinner for a complete response. The client is a React 18 SPA built with Vite and deployed on AWS Amplify.