A web service that finds people across escort and dating platforms by photo. Upload a picture — the system runs it through a neural network, compares face embeddings against a scraped database, and returns matching profiles with confidence scores.
The pipeline: a scraper continuously collects profiles from public sources, extracts faces, and stores vector embeddings. When a user uploads a photo, the system generates an embedding and runs approximate nearest-neighbor search across the database. Results come back with match confidence percentage and links to source profiles.
Built the face detection and recognition pipeline on dlib and OpenCV, with a Python backend handling the heavy lifting. The embedding database grew to hundreds of thousands of faces. Search took under a second even at that scale thanks to optimized vector indexing.