An Agentic AI System that transforms unstructured restaurant menu images into personalized, health-conscious recommendations.
- Multimodal Perception: Digitizes complex or handwritten menus using Vision LLMs.
- Personalization Engine: Automatically filters menu items based on user health profiles (e.g., Vegan, Keto, Diabetic-friendly).
- Data Integration: Designed to interface with Public Nutrition Databases (Government APIs) for real-time calorie and macronutrient tracking.
- Agent Workflow: Uses a reasoning-based approach (Think-Act-Observe) to ensure recommendations are logical and safe.
| Component | Technology |
|---|---|
| Language | Python 3.9+ |
| AI Model | OpenAI GPT-4o-mini (Vision & Reasoning) |
| Framework | Streamlit (Interactive UI) |
| Agent Logic | Built-in Agentic Workflow with Prompt Engineering |
| Data Handling | Pydantic & JSON |
- Input: User uploads an image of a menu and defines their dietary constraints.
- Analysis: The Vision model extracts menu items and descriptions.
- Augmentation: The Agent fetches nutritional data for the detected items.
- Reasoning: The system evaluates each dish against the user's profile.
- Output: A curated selection of dishes with nutritional breakdowns and reasoning.