mariocsilva / rag_qa_llm

Knowledge-Sharing Hub using RAG Q&A techniques with LLMs (Llama2 and ChatGPT)

RAG_Q&A_LLM

In this repository, I delve into creating a knowledge-sharing hub from my own data sources where teams can get insights and answers with the ease of a conversation, using the RAG Q&A technique, with the potential to complete the way information is shared within both small and large organizations.

In short, RAG Q&A (Retrieval-Augmented Generation Question and Answering) is a framework that blends retrieval-based and generation-based approaches. It employs a retriever to find relevant information and, typically, an LLM (Large Language Model) as the generator to create contextually rich answers. By definition, LLMs are a type of AI model capable of understanding and generating human-like language, trained on vast amounts of text data to learn the patterns, structures, and context of natural language.

I made one notebook using the free Llama-2–7b-chat-hf model (with a 4-bit quantization strategy) for Google Colab, which allows running this software for free, and another notebook using the OpenAI chat-gpt-3.5-turbo LLM model using an OpenAI subscription. Please make sure to install the requirements if you're going to use the local version.

For this use case, markdown documentation files were retrieved from three different data sources:

• Full-Stack App Documentation: Personal project well documented with markdown files.
• Short SQL Tables Descriptions: Repository documenting some short SQL tables.
• Demo Confluence Space: A public demo confluence space. The chatbot scrapes data in a markdown format.

These documents were chosen because they are typical ways for organizations to document and preserve their projects' knowledge.

The final chatbot extracted knowledge accurately, providing concise and relevant answers, while linking the documents it used.