- My role
- AI engineer on a team of five: refining the model, plus a little design
- Where
- UC Davis, team of five
- When
- Jan – Mar 2026
- Stack
- SwiftUI
- Flask (async)
- Firestore
- Claude API
- OpenAI API
- Computer vision
Munch, live on the App Store
Cook what you’ve got.
Built by a team of five at UC Davis, Jan to Mar 2026. I was the AI engineer: refining the model, plus a little design. Snap your fridge, and the app works out what’s inside and what you can make with it. Nothing is saved until you confirm.
The problem
Preflop
A full fridge and “nothing to eat.” Every recipe app starts from the recipe and sends you shopping.
The approach
Flop
Flip it: start from the pantry. Snap your fridge, Munch picks out what’s inside, and nothing is saved until you confirm.
The hard part
Turn
Scanning was the easy part. Pantry state, quantities, latency and taste were the real game: making it one product instead of a pile of AI demos.
What shipped
River
SwiftUI app on an async Flask API with 10+ REST endpoints, SMS auth and Firestore. Claude and OpenAI handle ingredient matching and recipe generation; every ingredient is checked against the pantry.
The result, or as poker players say, the showdown
Shipped to the App Store. “Cook what you’ve got.”