Welcome to the Old Recipes Finder, an application that brings the culinary wisdom of the past into the present through an interactive and user-friendly interface. This RAG-LLM Demo Project utilizes machine learning to process and retrieve historical recipes from a PDF book of old recipes from the Chicago Women"s Club of 1887.

Hugging Face Demo: https://huggingface.co/spaces/sarat2hf/RAG_LLM

The Old Recipes Finder application employs a multi-stage process involving data cleaning, processing, structuring, and retrieval augmented by language models (LLM) to deliver precise recipe information to users. The system relies on a vector database of recipe embeddings to match user queries with the most relevant historical recipes.

1. Data Source: It starts with a PDF book of historical recipes.
2. Data Cleaning: The content is standardized using LLM.
3. Data Processing: The clean data is prepared and summarized into manageable text windows, with embeddings generated for machine readability.
4. Data Structuring: Embeddings are stored in a JSON-formatted vector database.
5. RAG Stage: User queries are transformed into embeddings and matched with the database.
6. LLM Stage: The matched content is processed by an LLM to provide the final recipe output.

Before running the application, ensure you have the following prerequisites installed:

• Python 3.9+
• Flask
• Pandas
• NumPy
• Access to RAG and LLM APIs (Gemini by Google)