SPAI

Sterile processing is where surgical safety starts. Every instrument tray that reaches an operating room was cleaned, inspected and packed by a technician first. A missing glove, the wrong instrument or a contamination risk at that stage can follow the tray all the way into surgery.

Students, technicians and instructors need a second set of eyes that never gets tired.

I built SPAI on my own, from the models to the app.

• A glove, PPE and hand model. I combined existing YOLOv8 datasets and trained my own detector in Google Colab.
• An instrument and tray model. A second model trained on surgical instruments and trays, so the app can check what's actually on the tray.
• A visionOS app. Built in SwiftUI and RealityKit, it walks a technician through each prep step and shows detection alerts as overlays in their view.

The first version of my glove model, after 15 training epochs, correctly found 757 of 858 gloves in the validation set.

• Run the models on the headset itself with Core ML, so detection works without a server.
• Bring instrument and tray checks into the live app flow.
• Test with real sterile processing workflows and instructors.