RAG-nificent: An AI Chatbot powered by LLMs for Citation of Custom PDFs, Reports, and Guidelines

Supports Llama-3 by Meta AI

RAG-nificent is a state-of-the-art repository that leverages the power of Retrieval-Augmented Generation (RAG) to provide instant answers and references from a curated directory of PDFs containing information on any given topic such as WHO recommendations documents. This system is designed to aid researchers, policy makers, and the public in quickly finding specific information within extensive documents. Rag-nificent is powered by the Groq API, the fastes API (May 2024) for inference of LLMs resulting in (almost) instant responses.

Features

• Conversational Interface: Engage with the system using natural language queries to receive responses directly sourced from the PDFs.
• Direct Citation: Every response from the system includes a direct link to the source PDF page, ensuring traceability and verification.
• PDF Directory: A predefined set of key PDF documents, currently including WHO recommendations on major health topics such as schistosomiasis and malaria.

Available Models

• 📘 ChatGPT-3.5: Utilize this advanced iteration of the GPT model for engaging and human-like interactions, suitable for varied conversational tasks.
• 🦙 Llama3-70B-8192: Experience high-end performance with this large-scale model, ideal for complex language tasks and deep learning insights.
• 🦙 Llama3-8B-8192: Harness robust capabilities with this more accessible version of Llama3, perfect for a wide range of AI applications.
• 🌟 Mixtral-8x7B-32768: Leverage the power of ensemble modeling with Mixtral's extensive capacity for nuanced understanding and response generation.
• 🦙 Llama2-70B-4096: Utilize the proven effectiveness of Llama2 for comprehensive language processing and application development.
• 💎 Gemma-7B-IT: Explore specialized interactions and tech-focused solutions with Gemma, tailored for IT and technical content.

Demo

How It Works

The application utilizes a combination of OpenAI embeddings, Pinecone vector search, and a conversational interface to provide a seamless retrieval experience. When a query is made, the system:

1. Converts the query into embeddings.
2. Searches for the most relevant document sections using Pinecone's vector search.
3. Returns the answer along with citations and links to the source documents.

Setup

1. Clone the repository:

   git clone https://github.com/yourusername/RAG-nificent.git

2. Install dependencies:

   pip install -r requirements.txt

3. Set environment variables in a .env (also see .env.examplefile:

   • PINECONE_INDEX_NAME
   • PINECONE_NAME_SPACE
   • OPENAI_API_KEY
   • PINECONE_API_KEY
   • GROQ_API_KEY

4. Create a Pinecone index with the same name as PINECONE_INDEX_NAME. Set it up with dimensions=1536 and metric=cosine.

5. Place your PDFs in the pdf_data directory and run data_ingestion.py

6. Run the application:

   chainlit run src/app.py

Source Documents

The system currently includes guidelines from the following PDFs with direct links to the documents:

• WHO guideline on control and elimination of human schistosomiasis (2022)
• WHO guidelines for malaria (2023)