Advanced-RAG-LLM

This is the Back-end repo for the LLM API in the Advanced-RAG project encorporating feedback from the user

Installation

pip install -r requirements.txt

Run the server

python main.py

Embeddings with vector store

We convert our database schema into vector embeddings to match user queries with the respective tables. We have also incorporated a feedback-loop where user can give feedback on the selected tables and correct them.

API Endpoints

1. To get the related tables to a user query

   curl -X POST \
   http://localhost:5000/tables \
   -H "Content-Type: application/json" \
   -d '{"inputText": "how many clients are there?"}'

2. To get the response based on the choosen tables by the user

   curl -X POST \
   http://localhost:5000/response \
   -H "Content-Type: application/json" \
   -d '{"inputText": "how many clients are there?", "choosen_tables": ["table1", "table2"]}'

   The choosing of the tables is handled by the front-end logic

Change Models

By default, it will use OpenAI for both embeddings and inference. You can change the models_config object to change the models for embeddings and inference accordingly

models_config = {
  "llm": "gemini",
  "embeddings": "ollama"
}

This code creates the APIs to interact with our choice of LLM service. These APIs then interacts with NodeJS which acts as an interface between oue front-end and the LLM APIs.

• Rest of the NodeJS code is available here: Advanced-RAG-Node


• The front-end angular code is available here: Advanced-RAG-angular