FACEFIND
FaceFind is a privacy-first computer vision system for events. Attendees join with a short code, scan their face, and retrieve only the photos they appear in from a connected Google Drive album, without ever seeing anyone else's.
Built for events where the album is bigger than your patience. Hundreds of photos, thousands of faces, and one question every attendee asks: which ones am I in?
Face search on real event photos is hard: group shots, side profiles, bad lighting. It is also sensitive, because a face embedding is biometric data. The system had to be accurate and private at the same time.
YuNet detects faces, an ArcFace ResNet-100 ONNX model produces embeddings, and FAISS runs event-scoped vector search behind FastAPI, PostgreSQL, Redis, and React. Embeddings are AES-256-GCM encrypted, and it runs on Azure via Docker Compose and GitHub Actions, with 3,252 faces indexed across 299 files.
YuNet detects faces, an ArcFace ResNet-100 ONNX model produces embeddings, and FAISS runs event-scoped vector search behind FastAPI, PostgreSQL, Redis, and React. Embeddings are AES-256-GCM encrypted, and it runs on Azure via Docker Compose and GitHub Actions, with 3,252 faces indexed across 299 files.
A semester project taken all the way to a hardened production deployment, not a notebook demo.
An AI-powered lecture note-taking app that catches every word in real time, in Urdu and English, so nothing gets lost.
