LocalGraph RAG

A Retrieval-Augmented Generation (RAG) system built for offline, privacy-focused document analysis. It utilizes a stateful agentic workflow to retrieve, filter, and generate answers based on PDF context, running entirely on local hardware via Ollama.

Architecture

• Frontend: Streamlit
• Backend: Flask
• Orchestration: LangGraph (Stateful Agent Workflow)
• Vector Store: ChromaDB
• Inference: Ollama (Phi-3.5 / mxbai-embed-large)

Project Structure

• app.py: Frontend interface (Streamlit).
• server.py: Backend API (Flask). Handles ingestion pipeline and RAG state graph.
• colab/: Contains the original research notebook for cloud-based execution.

Setup

Prerequisites

• Python 3.10+
• Ollama (running locally)

Installation

1. Clone & Environment

   git clone <repo_url>
   cd chatbot_colab
   python -m venv venv
   source venv/bin/activate  # Windows: venv\Scripts\activate

2. Dependencies

   pip install -r requirements.txt

3. Model Setup Ensure Ollama is running and pull the necessary weights:

   ollama pull phi3.5
   ollama pull mxbai-embed-large

Usage

1. Backend: python server.py Server runs on port 8000.
2. Frontend: streamlit run app.py Interface available at http://localhost:8501.

Research & Prototyping

The core logic was originally validated in a Google Colab environment. If you do not have a local GPU capable of running LLMs, or wish to review the initial research phase, refer to the notebook:

• Notebook: colab/chatbot_work_proj.ipynb