zepto-labs/zepiris — reverse-engineered prompt

Build me a face authentication service like this repo describes. I want an API where a company can enroll people with a selfie, store their face record by tenant, and later search a new selfie to see who it matches. It should work for attendance, onboarding, or access control without OTPs or passwords.

Please include the full pipeline, face detection and embeddings, vector similarity search, image storage, and checks for blurry photos, spoofed faces, and unsafe images. Keep the API simple with endpoints to insert, upsert, delete, get, and search face records. Every response should include a request id so issues are easy to trace.

Set it up as two services, one main API and one ML inference service, with Docker Compose so I can run everything locally in one command. Include health checks, Swagger docs, sensible config through env vars, and basic tests. Use the current docs online if needed, but keep it practical and production ready enough to demo.