Full-stackJun 2026 – Present
Taro
Describe what you feel like eating and get a real recipe back, with macros, a grocery checklist, and a feed you can remix from.
Recipe generators mostly produce prose you then have to parse yourself. Taro enforces a JSON schema on the model's output with retry handling for malformed responses, so what comes back is structured ingredients, steps, macronutrients, and a checklist-format grocery list that drops straight into the UI.
The part I like most is remixing: any dish you are shown, whether a friend sent it or you found it in the feed, can be regenerated against your own constraints. Someone else's carbonara becomes your dairy-free carbonara without anyone typing a new prompt.
Currently in private beta.
What it does
- Built a full-stack social cooking platform generating recipes from natural-language prompts with schema-constrained output and retry handling.
- Returns structured ingredients, steps, macronutrients, and a checklist grocery list from a single prompt.
- Implemented a remix flow that regenerates any received or discovered dish against the user's own constraints.
- Social feed with sharing and recipe discovery, backed by Supabase and Auth0.
Built with
- Next.js
- TypeScript
- Supabase
- Auth0
- Kimi K3
- Azure AI Foundry