Iris Biometric Pipeline

The client needed iris recognition integrated into an existing Laravel web application. No off-the-shelf model existed for their iris dataset quality and capture conditions. The pipeline needed to be trained from their data and deployed as a latency-tolerant API.

Two models were trained from scratch: YOLO for iris detection, locating the eye region in camera input; UNet for segmentation, isolating the iris from sclera and eyelid. The FastAPI microservice wraps both behind a single endpoint and returns a feature vector the Laravel application stores and compares. Docker handles environment parity between development and the production server.