UMAMI is an AI-powered culinary assistant designed to make recipe recommendations more reliable and personalized. Unlike traditional AI chatbots that rely primarily on a language model, UMAMI grounds every response in a structured recipe database using a LangChain SQL Agent. The system integrates more than 50,000 recipes with allergen and ingredient substitution data, enabling users to search recipes naturally while receiving accurate recommendations, allergy-aware suggestions, nutritional information, and ingredient alternatives. The project demonstrates the complete data pipeline—from data collection and database design to AI integration, deployment, and visualization.
- [🍽️ Recipe recommendations]
- [
⚠️ Allergen-aware suggestions] - [🔄 Ingredient substitutions]
- [🥗 Nutrition information]
- [🤖 Reliable AI responses powered by SQL]
The raw data consisted of three datasets:
- The first dataset recipes containing recipes, ingredients, instructions and nutritional information.
- The second dataset allergenic foods containing a list of food allergens.
- The second dataset substitute ingredients containing a list of posible substitute ingredients.
| Data Engineering | AI | Deployment | Visualization |
|---|---|---|---|
| Python | GPT-4o-mini | Streamlit | Tableau |
| Pandas | LangChain SQL Agent | SQLite | |
| SQLite | OpenAI API | GitHub |
Build a SQL agent, LangChain Docs: https://docs.langchain.com/oss/python/langchain/sql-agent
Advanced-RAG-Series: https://github.com/Farzad-R/Advanced-QA-and-RAG-Series
Allergenic Food Database, University of Manchester: http://research.bmh.manchester.ac.uk/informall/allergenic-foods/
Ingredient Substitution: https://www.ndsu.edu/agriculture/extension/publications/ingredient-substitutions