I realized there was a cold start problem so I added a functionality to start from a fridge photo.

You can try the app without registering, but the agent can keep track of preferences of the user of registered.

I built this MVP using Lovable. As a dev, I wanted to see how far I could push "AI-assisted development" for a full-stack PWA. It allowed me to skip the boilerplate and focus entirely on the prompt-engineering for the Gemini vision logic and the Supabase integration. I was skeptical about "generated code," but the speed at which I could iterate on the UI while maintaining a clean React/Supabase architecture was a game-changer for this specific project.