Mcprag

🚀 mcpRAG

A Retrieval Augmented Generation (RAG) system built with open - source embeddings, an open - source vector database, and Gemini LLM.

🚀 Quick Start

In this project, I've created a text - file - based RAG system:

• Embeddings: I used "nomic embeddings" and ran them locally via ollama.
• Large Language Model (LLM): I adopted the Gemini - 2.0 - flash version.
• Vector Database: I utilized FAISS.

All text files are chunked. Each chunk contains the file name, chunk ID, and chunk text, and is stored locally in JSON format. Each chunk is converted into an embedding and collected into a list.

These embedding lists are indexed using FAISS and stored locally. When a query is made, nomic embeddings are used to generate an embedding for the query content, and relevant positions (chunk indices) are searched for in the FAISS index. These indices are then used to retrieve the actual text from the JSON files.

The retrieved text is then passed to the LLM along with the query to generate an answer. Additionally, extra text can be appended to the existing index, and new queries can be run on the updated index by loading the stored index and embedding files.

✨ Features

• Open - source Components: Leverages open - source embeddings and a vector database, reducing costs and increasing flexibility.
• Local Operation: Allows local running of embeddings, enhancing data privacy and reducing dependency on external services.
• Index Updatability: Supports appending extra text to the existing index and running new queries on the updated index.