COMPUTER VISION
AutoInspect AI — Vehicle Damage Detection
Production vehicle damage classifier using ResNet50 with transfer learning. Detects 6 damage categories from uploaded car images via a FastAPI backend hosted on HuggingFace Spaces.
Public demo
6 — damage categories detected
The problem
Sorting vehicle-damage photos by damage type is a repetitive first step in a claim. The project builds a classifier for that step together with the serving, monitoring and interface around it — an exercise in taking a vision model all the way to a deployed product.
The approach
ResNet50 fine-tuned via transfer learning on a multi-class vehicle damage dataset covering 6 damage types. FastAPI backend hosted on HuggingFace Spaces exposes a REST endpoint; Next.js 16 App Router frontend handles image upload, Prometheus metrics, and a live analytics dashboard built with Recharts.
The results
Classifies Front Breakage, Rear Breakage, Door Dent, Glass Shatter, Head Lamp Damage, and Tail Lamp Damage. Live Prometheus monitoring tracks API health, request counts, and average inference latency. Dashboard shows per-category prediction confidence breakdowns.
Stack: Next.js 16, TypeScript, ResNet50, FastAPI, HuggingFace Spaces, Recharts