Decade-qiu/CookHero

What it solves

CookHero is a personalized diet management platform that transforms cooking and health goals into actionable daily routines. It solves the difficulty of planning healthy meals, tracking nutritional intake from text or images, and finding professional cooking guidance tailored to specific dietary restrictions or health goals (such as weight loss or blood sugar control).

How it works

The system combines several advanced AI architectures to manage nutrition and cooking:

• Agentic Workflow: Uses a ReAct (Reasoning and Acting) loop with a "ToolHub" and a system of specialized sub-agents to make autonomous decisions and call tools for planning, searching, and analysis.
• Hybrid RAG: Employs a combination of vector search (Milvus), keyword search (BM25), and a Reranker (e.g., Qwen3-Reranker) to retrieve accurate recipes from a global library and user-uploaded personal cookbooks.
• Multimodal Processing: Integrates vision models to identify ingredients and dishes from uploaded images, which then feeds into the diet recording and nutritional estimation system.
• Nutritional Tracking: Automatically calculates calories and macronutrients from AI-parsed diet records and compares them against user-defined goals.
• Infrastructure: Built on a full-stack architecture using FastAPI, React, PostgreSQL, Milvus, and Redis.

Who it’s for

• Kitchen beginners seeking professional guidance.
• Individuals focused on fitness, weight loss, or blood sugar management.
• People with food allergies or specific dietary requirements.
• Health-conscious users wanting a structured way to plan and track their weekly meals.

Highlights

• Intelligent Diet Manager: An AI agent that handles meal planning, nutritional analysis, and real-time web searching via Tavily.
• Automated Diet Logging: One-click recording of meals via text or images with automatic calorie and macronutrient estimation.
• Expert Sub-agent System: Supports both built-in and user-defined expert agents with independent system prompts.
• RAG Quality Control: Integrated RAGAS framework to monitor faithfulness and answer relevancy.
• Security Layer: Implements prompt injection protection, rate limiting, and structured audit logs using NeMo Guardrails.
• MCP Support: Compatible with the Model Context Protocol for extending tools, including integrations like Amap.