RAGademic: Query Your University Notes and Books!

This project provides a Retrieval-Augmented Generation (RAG) system to query your university notes and books efficiently (could be used to retrieve any kind of document really, but this is the use case i've opted for! :) ). It allows users to upload PDF documents, store them in a Chroma vector database, and interact with them through an LLM-powered chatbot. The project also includes a visualization tool for document embeddings.

Features

• Chat with your notes: Ask questions, and the system retrieves relevant information using embeddings.
• Database management: Upload, delete, and visualize document embeddings.
• Arxiv source addition: Search for papers on specific topics and add them to your knowledge base.
• Three LLM options: Use either OpenAI's API or local models like LLaMA3.2 and the new Gemma3 via ollama chat.
• Three Embedding options: Use either OpenAI's embedder, HF embedder from the hub or chroma builtin local embedder.
• Persistent memory: Keeps track of conversations in between sessions, using llama3.2 as a summarizer to keep context windows occupation optmized.
• Web Search Agent: If activated, it evalutaes (using local llama3.2) the pertinence of the local knowledge wrt the query of the user and if necessary it scrapes the web for more context.

---

Installation

Prerequisites

• Ensure you have Python 3.10+ installed.
• Install Ollama to run local models.

1. Clone the Repository

git clone https://github.com/yourusername/ragademic.git
cd ragademic

2. Create Env end install requirements

python3 -m venv ragademic
pip install -r requirements.txt

Also create .env file and add:

OPENAI_API_KEY=<your_key>
EMBEDDER_TYPE=<type> 
HF_TOKEN=<your_token> 
MODEL_SIZE=<number_of_params>

Notes:

• you can use llama3.2 installing it locally with ollama.
• the openai api key is needed if you use gpt for chatting and/or openai embeddings.
• embedder_type can be huggingface or openai or local which uses chroma built in vectorizer.
• the huggingface embedding is obtained through API but it could be done locally too.
• when using Gemma3, provide the size of the model you pulled.

3. Load your PDF

Load PDFs

./knowledge_base
    /category_1/*.pdf
    ...
    /category_n/*.pdf

4. Run the app

python3 app.py

Chat Interface

DB Management Interface

Embeddings Visualization

Author:

Gianmarco Russo