---
title: Explainable & Auditable Alloy Defect Detection
emoji: 🔍
colorFrom: gray
colorTo: red
sdk: gradio
sdk_version: 5.31.0
app_file: app.py
pinned: false
---

# Explainable & Auditable Alloy Defect Detection

Group 8's submission for Problem #10 (MIT Professional Education, "Applied AI for
Materials Discovery," July 2026).

Upload a steel-surface micrograph and see: the classifier's prediction, a live
Grad-CAM heatmap explaining it, the deletion-curve faithfulness test that checks
whether the heatmap is trustworthy, and the Conformal Risk Control guarantee that
gives a provable — not just calibrated — bound on error rate.

Full notebook, one-page summary, and slides in the companion project folder.