SmartMinutes

This project is a complete system for summarizing, and extracting relevant information from meetings using Large Language Models (LLMs) running locally. The system supports auxiliary documents to enhance Meeting Minutes accuracy by employing the Retrieval-Augmented Generation (RAG) technique.

The source code freely available under the MIT license.

Contents

• SmartMinutes
• Contents
• Initial information
• Hardware Requirements
• Software Requirements
• Downloading the LLM
• Quick Start
  • Clone the Repository
  • Backend Setup
  • Frontend Setup
• Project Structure
• Configuration
• Project Structure
• Features
• Contributing

🍭 Initial information

The project is divided into two main components:

• Backend: Handles transcription processing, summary generation, and AI model integration.
• Frontend: A simple and efficient static web interface that allows users to upload transcriptions, view summaries, transcriptions and manage auxiliary documents.

🖥️ Hardware Requirements

The system has been tested on the following hardware configuration, and a similar setup is recommended for optimal performance:

• Processor: AMD Ryzen 5 3600 or higher;
• RAM: 16GB (minimum);
• Graphics card: RTX 3060 12gb;
• Storage: NVMe SSD recommended with at least 15GB of free space;

🛠️ Software Requirements

Before running the project, ensure the following dependencies are installed:

• Operating System: Linux (tested on Ubuntu 22.04) or Windows 10
• Python: Version 3.11 or later
• Ollama: Required for running LLMs locally, version 0.3.12 (Installation Guide)

🔽 Downloading the LLM

The project is optimized for gemma-2-9b-it-sppo-iter3, which can be downloaded via Ollama with the following command:

# Downloads the LLM model by using Ollama
ollama pull lucasalmeida/gemma-2-9b-it-sppo-iter3:Q4_K_M

🚀 Quick Start

1. Clone the Repository

git clone https://github.com/lucasandre22/smart-minutes.git

2. Backend Setup

Navigate to the backend directory and create a virtual environment:

cd backend
python3 -m venv venv
source venv/bin/activate  #On Windows: venv\Scripts\activate
pip install -r requirements.txt

Run the backend server:

uvicorn api.main:app --host=0.0.0.0 --port=8000

If preferable, run the backend server with Hot Reload:

uvicorn api.main:app --host=0.0.0.0 --port=8000 --reload

3. Frontend Setup

Just kidding, there are no action to be taken to setup the frontend :).

The application should now be accessible in your browser at http://localhost:8000.

📂 Project Structure

├── backend/            # Backend service (FastAPI, LLM processing, RAG)
│   ├── action_items/   # All the classes related to the action items feature
│   ├── api/            # Classes related to the FastAPI
│   ├── core/           # config.py file and RAG database loading method
│   ├── custom_request/ # Classes related to the Custom Request feature
│   ├── databases/      # Where all the files consumed by the system are stored, like transcripts (Created when starting the project for the first time)
│   ├── llms/           # The standard LLM classes, used through the code
│   ├── minutes/        # Classes related to the Meeting Minutes method
│   ├── rag/            # Classes related to RAG
│   ├── scripts/        # File parsers scripts
│   ├── summarization/  # Classes related to the Summarization method
│   ├── tests/          # Test classes
├── frontend/           # Frontend application (Bootstrap)
│   ├── assets/         # UI assets (icons)
│   ├── css/            # CSS Stylesheets file
│   ├── js/             # Javascript files
│   └── *.html...       # Static HTML pages
├── docs/               # Documentation and design assets
└── README.md           # Project documentation

🔧 Configuration

You can configure environment variables in a backend/core/config.py file for backend, including API keys and model settings.

Also, you can navigate to the settings page in the applications and configure:

📖 Documentation

The full project documentation is available in the docs/ folder. It includes:

• System Overview: Architecture and main functionalities.
• API Reference: Endpoints and usage examples.

To view the documentation, navigate to the localhost:8000/docs endpoint.

📌 Features

✅ AI-powered summarization, action items, meeting-minutes and custom requests about the meeting ✅ Support for external documents (RAG) ✅ Local LLM execution (no external API costs)
✅ Web-based user interface

📖 Final considerations

• The prompts for every technic follows the DeepEval architecture, it is defined in template_refine.py files. Example: all the prompts regarding the minutes, are defined in backend/minutes/template_refine.py minutes module.
• All the LLM prompts are in Portuguese. The idea is to have English in the future versions too.

🤝 Contributing

Feel free to submit issues, pull requests, or suggestions to improve the system!