01 / The problem
What the system needs to solve
Most sellers have no reliable way to price scrap: they don't know which material they have, how much of it there is or what it's worth. The app has to turn photos into a credible estimate and a listing buyers can act on.
02 / The approach
How the pieces fit together
- Identify materials from multiple photos with YOLOv8 and TensorFlow Lite models.
- Estimate size and weight, then calculate a market value automatically.
- Serve analysis, history and listings from a Node.js and Express API backed by PostgreSQL, Prisma and Redis.
- Build the Android client in Kotlin with Jetpack Compose and secure authentication.
03 / Key features
What it does
- Multi-image AI analysis
- Automated price calculation
- Personalized analysis history
- Digital marketplace for listing items
04 / The tradeoffs
Engineering is a set of choices
Photo-based estimates depend on lighting, angle and scale, so a single image is not enough. Multi-image analysis costs more processing per item but gives the valuation more evidence to work from.
05 / Validation considerations
What to test before relying on it
Compare estimates against weighed and priced samples across material types and photo conditions before relying on valuations.