LLM-RAG API

A Flask-based application providing multiple endpoints for performing vector searches and Retrieval-Augmented Generation (RAG) using MongoDB and LangChain. The application leverages embeddings for document retrieval and query augmentation.

Table of Contents

• LLM-RAG API
  • Table of Contents
  • About the Project
  • Features
  • Technologies Used
  • Software Architecture
  • Project Structure
  • Setup and Installation
  • Usage
    • Hello World
    • Vector Search
    • RAG
    • Self-Querying Vector Search
    • Self-Querying RAG
  • Example
  • Contributors

About the Project

Course: Open Learning
Semester: 6

This project was developed as part of coursework in Open Learning. It aims to provide a scalable API for vector search and RAG tasks using MongoDB, Flask, and LangChain, facilitating efficient document retrieval and query augmentation.

Features

• Vector Search
• Retrieval-Augmented Generation
• Self-Querying Vector Search
• Self-Querying Retrieval-Augmented Generation

Technologies Used

• Programming Languages/Technologies:
  • Python
• Libraries/Frameworks:
  • Flask
  • LangChain
• Database:
  • MongoDB
• Tools:
  • Large Language Models (LLMs)

Software Architecture

The application is designed as RESTful microservice with endpoints for vector search and RAG. MongoDB is used for data storage, and LangChain facilitates interaction with language models and embeddings.

Project Structure

llm-rag-api/
│
├── misc/               # Helper functions
│
├── rag/                # Customised RAG implementation
│
├── README.md           # Project documentation
├── app.py              # Flask API endpoints
└── requirements.txt    # List of dependencies

Setup and Installation

Follow the instructions below to set up the project on your local machine.

1. Clone the Repository:

   git clone https://github.com/grzgm/llm-rag-api
   cd llm-rag-api

2. Install Dependencies:

   pip install -r requirements.txt

3. Configure Environment Variables: Copy the contents of .env.example to a .env file in the project root and populate it with the necessary environment variables:

   MONGO_URI=<your_mongodb_uri>
   DB_NAME=<your_database_name>
   COLL_NAME=<your_collection_name>
   INDEX_NAME=<your_index_name>
   EMBEDDING_KEY=<your_embedding_key>
   DOCUMENT_CONTENT_DESCRIPTION=<document_content_description>
   

4. Run the Flask Application:

   flask run

Usage

Hello World

• URL: /
• Method: GET
• Description: Returns a simple greeting message to ensure the server is running.
• Parameters: None
• Response:

  <p>Hello, World!</p>

Vector Search

• URL: /vector-search
• Method: GET
• Description: Performs a vector search on the MongoDB collection.
• Parameters:

Parameter	Type	Required	Description
query	string	Yes	The query to search for.
projection	JSON string	No	Custom projection for the MongoDB query.
docs_num	integer	No	Number of documents to return (default: 3).

• Response: JSON array of documents matching the search query.

RAG

• URL: /rag
• Method: GET
• Description: Performs a RAG operation to retrieve documents and generate responses based on the query.
• Parameters:

Parameter	Type	Required	Description
query	string	Yes	The query to search for.
projection	JSON string	No	Custom projection for the MongoDB query.
docs_num	integer	No	Number of documents to return (default: 3).

• Response: JSON object containing the RAG-generated response.

Self-Querying Vector Search

• URL: /sq-vector-search
• Method: GET
• Description: Performs a self-querying vector search, which automatically generates filters based on the query.
• Parameters:

Parameter	Type	Required	Description
query	string	Yes	The query to search for.
projection	JSON string	No	Custom projection for the MongoDB query.
docs_num	integer	No	Number of documents to return (default: 3).

• Response: JSON array of documents matching the search query with auto-generated filters.

Self-Querying RAG

• URL: /sq-rag
• Method: GET
• Description: Performs a self-querying RAG operation to retrieve documents and generate responses based on the query with auto-generated filters.
• Parameters:

Parameter	Type	Required	Description
query	string	Yes	The query to search for.
projection	JSON string	No	Custom projection for the MongoDB query.
docs_num	integer	No	Number of documents to return (default: 3).

• Response: JSON object containing the self-querying RAG-generated response.

Example

To use the endpoints, send HTTP GET requests with the appropriate parameters to the Flask server. For example, to perform a vector search, use the following curl command:

curl -G \
  --data-urlencode "query=your_search_query" \
  --data-urlencode "projection={\"field\": 1}" \
  --data-urlencode "docs_num=5" \
  http://localhost:5000/vector-search

Make sure to adjust the parameters as needed.

Contributors

• Grzegorz Malisz: Author.