Face Lab

Small face checks, no ceremony.

Add a person, find a person, or compare two photos. The app keeps the math visible enough to trust.

How the app reads a face

  1. The detector finds one face in the photo and ignores the rest of the scene.
  2. The model turns that face into 128 numbers called an embedding.
  3. The app compares distances. Close numbers usually mean the same person. Far numbers usually mean no.
  4. A distance below 0.55 counts as a match for this capstone build.

What is inside

The backend uses FastAPI, dlib, and face_recognition. The frontend is Astro. The point is simple: show the decision, the distance, and the confidence without hiding the rough edges.

Best photos

Front facing, bright, and uncropped. Sunglasses and motion blur make the model guess harder.