🎙️ Multilingual Meeting Assistant Backend

This is a robust FastAPI-based backend designed to power the Multilingual Meeting Assistant. It streamlines the lifecycle of meeting documentation—from raw audio ingestion and high-accuracy transcription to intelligent summarization and professional PDF report generation.

The system now supports a hybrid architecture, offering both cloud-based processing via Alibaba Cloud and a fully local, open-source pipeline (Whisper + HuggingFace) for INDIAN users without cloud console access.

---

✨ Key Features

• Dual Transcription Engine: Supports Alibaba Cloud ASR for enterprise-grade speed or a local Whisper-FFmpeg implementation for privacy and cost-efficiency.
• Intelligent Summarization: Leverages LangChain integrated with Qwen/DeepSeek models via HuggingFace pipelines to extract action items and key takeaways.
• Professional Documentation: Automated generation of meeting notes in PDF format.
• Comprehensive API: RESTful endpoints for meeting management (CRUD), advanced search within transcripts, and file exports.

---

🛠️ Tech Stack

Component	Technology
Framework	FastAPI (Python 3.8+)
ORM/Database	SQLAlchemy, SQLite/PostgreSQL
AI/ML (Cloud)	Alibaba Cloud ASR, Qwen LLM
AI/ML (Local)	OpenAI Whisper, HuggingFace Transformers
Orchestration	LangChain
Processing	FFmpeg (Audio), Pydantic (Validation)

---

🚀 Getting Started

📋 Prerequisites

• Python 3.8 or higher.
• Optional: Alibaba Cloud account with ASR enabled (if using the cloud method).
• System Level: FFmpeg installed on your machine.

📥 Installation

1. Clone the Repository

git clone https://github.com/your-username/meeting-assistant-backend.git
cd meeting-assistant-backend

2. Environment Setup Create a virtual environment and activate it:

python -m venv venv
# Windows:
venv\Scripts\activate
# macOS/Linux:
source venv/bin/activate

3. Install Dependencies

pip install -r requirements.txt

4. Configuration Create a .env file in the root directory:

ALIBABA_ACCESS_KEY_ID=your_id
ALIBABA_ACCESS_KEY_SECRET=your_secret
ALIBABA_APP_KEY=your_app_key

5. Run the Application

uvicorn main:app --reload

Access the interactive API docs at: http://localhost:8000/docs

---

🌍 Alternative Method (Local/No API Keys)

For users unable to access the Alibaba Cloud Console (including users in specific regions like India), follow this setup to run the Whisper-FFmpeg and HuggingFace Pipeline. This method runs entirely on your local hardware.

1. Install GPU-Accelerated PyTorch (Recommended)

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

2. Install AI Libraries

pip install git+https://github.com/openai/whisper.git
pip install transformers accelerate langchain huggingface_hub ffmpeg-python einops==0.7.0

3. Model Selection

By default, the system is configured to use Qwen1.5-0.5B (~1GB) for summarization, which offers a great balance between performance and resource usage. You may also opt for the smaller Qwen-2 models.

---

📂 Directory Structure

├── main.py                # Application entry point
├── database.py            # Database connection logic
├── models.py              # SQLAlchemy database models
├── schemas.py             # Pydantic data schemas
├── services/              # Core business logic
│   ├── transcription.py   # Alibaba & Whisper integration
│   ├── summarization.py   # LangChain & HuggingFace logic
│   └── pdf_service.py     # Report generation logic
├── uploads/               # Raw audio storage
└── pdfs/                  # Generated meeting notes

---

📡 API Reference

Method	Endpoint	Description
POST	/meetings/	Upload and process a new recording
GET	/meetings/	List all processed meetings
GET	/meetings/{id}	Retrieve details of a specific meeting
DELETE	/meetings/{id}	Remove a meeting record and files
POST	/meetings/{id}/export	Generate and download PDF notes
GET	/meetings/{id}/search	Query the transcript for specific terms

---

☁️ Deployment Tips

When deploying to platforms like Render, Railway, or a VPS:

1. Dockerize the App: Use a Dockerfile to ensure a consistent environment containing Python, FFmpeg, and the required Cuda drivers.
2. System Binaries: Ensure ffmpeg is installed at the system level (apt install ffmpeg).
3. Dependency Management: Ensure openai-whisper and ffmpeg-python are explicitly listed in your requirements.txt.
4. Local Testing: Always test your containerized build locally to verify that Whisper can access the system path for FFmpeg.

---