RAG demo using Langchain , FAISS, LLAMA3 and Streamlit

Introduction

The application illustrates use of Langchain retrievers, generators and prompt templates to create a RAG pipeline LLAMA3 8B is used as the LLM to process the input query

Design

The code is structured into 4 components :

1. App/The driver code : Has the streamlit calls, along with instantiation of other components. The chat history is also preserved here in memory
2. The Retriever : This creates the loader for various input formats such as documents,confluence, webpages etc converts the doccuments into word embeddings and then stores the same in a vector database for quick retrieval
3. Generator : The generator is reponsible for creating the prompt using the chat history, query and data from retriever.
4. LLM interface * : Currently kept in the most basic format , this will house the LLM APIs, access tokens etc.

How to use

1. pip install -r requirements.txt
2. Put relevant documents in data folder
3. change app code to point to the appropiate path in the data folder
4. streamlit run app.py

Notes

1. Confluence links are removed for security reasons, you can use the confluence links of your organization to get it working with confluence data
2. Disclaimer : This is still work in progress , so use with caution ;)

Authors

Ankur Neog /[email protected]