I’m spinning up a lightweight backend that lets a client upload an image, quietly passes that image to Google’s Gemini Vision model, and ships back a clean, well-structured JSON payload. I’ve settled on Gemini as the primary vision API, but if you have a compelling reason to introduce an alternate provider later, I’m open to discussing it. Here’s what I need built: • A REST endpoint (multipart/form-data) for image uploads. . Using supabase to handle uplaods • Seamless hand-off of the file to the Gemini Vision API. • Parsing of Gemini’s response so the client receives tidy, predictable JSON keys and values. • Basic but rock-solid validation: file type, size, and any other quick checks you normally wire in to keep the service safe. • Sensible error handling—clear status codes and helpful messages. • Simple user authentication so only authorised callers can hit the upload endpoint (token-based is fine). Tech stack is flexible between Python/FastAPI and Node.js/Express; FastAPI is my default choice because of its built-in docs and async flow, but feel free to pitch Express if that’s your home turf. Deliverable: a small repo I can run with Docker-Compose or a straightforward README, plus any environment variables I’ll need for Gemini credentials. A tiny test script or Postman/Insomnia collection showing the happy path would be perfect for acceptance.