HelloFresh Recipe Recommendation Engine with AI Agent

• Automated Menu Scraping: Fetches the current week's HelloFresh menu dynamically.
• Recipe Data Extraction & Structuring: Extracts detailed recipe information including description, ingredients, utensils, instructions, and nutrition.
• Vector Database Integration: Stores recipe embeddings in Qdrant for efficient similarity search and recommendations.
• Traditional Database Storage: Saves original recipe data in an SQLite database for comprehensive record-keeping.
• AI Agent for Chat Interface: Enables users to interact with an AI agent to get personalized recipe recommendations.
• Qdrant Recommend API: Utilizes Qdrant's advanced recommendation capabilities, accepting positive and negative preferences.
• Metadata Enrichment: Enriches recipe data with relevant metadata like week, cuisine, category, and tags.

Workflow Steps:

1. Fetch Available Courses: The Get This Week's Menu node fetches the current week's HelloFresh menu URL. The Extract Server Data node then extracts the embedded JSON data from the page. Finally, the Extract Available Courses node parses this JSON to get a list of available recipes.
2. Create Recipe Documents for VectorStore: For each available recipe, the workflow:
   • Makes an httpRequest to get the full recipe page (Get Recipe).
   • Extracts detailed recipe information using CSS selectors in the Extract Recipe Details node.
   • Combines the course metadata (Get Course Metadata) and recipe details (Extract Recipe Details) using the Merge Course & Recipe node.
   • Formats this combined data into a structured document for the vector store and database in the Prepare Documents node.
3. Vectorise Recipes for Recommendation Engine:
   • The Save Recipes to DB node stores the prepared recipe documents in an SQLite database.
   • The Qdrant Vector Store node then embeds these documents and inserts them into the Qdrant vector database, creating embeddings for recommendation.
4. Chat with HelloFresh Recommendation AI Agent:
   • The Chat Trigger node initiates the AI agent interaction.
   • The AI Agent node handles the conversational flow, understanding user preferences.
   • When the agent needs to find recipes, it triggers the Qdrant Recommend API node.
   • The Get Mistral Embeddings node is used to generate embeddings for the user's positive and negative preferences.
   • The Use Qdrant Recommend API node then queries Qdrant with these embeddings and filters by the current week.
   • The Get Recipes From DB node retrieves the full recipe details from the SQLite database for the recommended recipe IDs.
   • The AI agent uses this information to formulate a recommendation to the user.