Markdown

🎙️ VideoSUM Workspace

An advanced, locally-driven multimedia intelligence workspace designed to ingest online broadcasts (YouTube URLs) and local recording containers, execute parallelized speech-to-text transcription matrix streams, and extract synthesized executive meeting insights via Cloud LLM pipelines. Featuring a native Retrieval-Augmented Generation (RAG) system to query long-form document vector spaces directly from an interactive user interface.

---

🚀 Architectural Blueprint

The application architecture follows a highly optimized, decoupled data processing pipeline:

[YouTube URL / Local File] │ ▼ (pydub / ffmpeg wrapper) [Audio Track Ingestion & 10-Min Slicing Chunks] │ ▼ (Local Whisper Engine CPU Matrix) [Raw Text Log Transcription Node] │ ├───► [Mistral Cloud LLM Core] ───► [Executive Minutes / Summaries Report] │ ▼ (ChromaDB Vector Indexing Pipeline) [Persistent RAG Embeddings Node] ◄───► [Interactive Chat UI Playground Input]

---

✨ Features

• Decoupled Ingestion Node: Toggle fluidly between streaming media from YouTube URLs or uploading heavy local media containers (.mp3, .wav, .mp4, .mkv, etc.).
• Smart Slicing Core: Auto-slices lengthy audio into uniform 10-minute structural segments using pydub to prevent local processing pipeline choke points.
• Local Acoustic Core: Runs local multi-threaded Whisper transcription matrices directly over your CPU framework.
• Cross-Lingual Transformation Node: Integrated translation layer to convert regional audio tracks (e.g., Hindi speech elements) cleanly into English logs.
• Executive Synthesis Report Engine: Leverages powerful cloud LLMs (Mistral AI) to process long-form text logs into polished, professional meeting minutes and summary records.
• Document RAG Node: Indexes your transcription assets into a localized vector document store (ChromaDB), unlocking real-time semantic query lookups.

---

🛠️ Installation & Setup Workspace

This project leverages uv, the fast Python package installer and resolver ecosystem.

1. Prerequisites

Ensure you have system level ffmpeg installed and mapped to your global environment paths.

2. Clone and Navigate to Directory

git clone [https://github.com/devthegod1/VideoSUM.git](https://github.com/devthegod1/VideoSUM.git)
cd VideoSUM
3. Initialize the Virtual Environment via uv
Bash
# Create a localized virtual environment environment node
uv venv

# Activate the workspace virtual environment node
# On Windows PowerShell:
.venv\Scripts\Activate.ps1
# On Linux/macOS:
source .venv/bin/activate
4. Install Project Packages
Bash
uv pip install -r requirements.txt
5. Configure Your Private API Credentials Matrix
Create a secret environment configuration file named .env inside the root repository tracking space (Note: This file is blocked by .gitignore and stays secure on your machine):

Code snippet
MISTRAL_API_KEY=your_secret_mistral_cloud_api_key_goes_here
🎮 Execution Management
To launch the high-contrast analytical workspace panel layout engine, execute the following command:

Bash
streamlit run app.py
Open your local browser gateway port (typically http://localhost:8501) to interact with the ingestion nodes.

📂 Project Directory Topography
Plaintext
VideoSUM/
├── data/                   # Audio track containers, cache maps, & database shards (Ignored)
├── src/                    # System Modular Pipeline Logic Sub-directories
│   ├── __init__.py
│   ├── audio_ingestion.py  # Audio stream downloads, track file saving, and slicing
│   ├── transcription.py    # Local Whisper CPU execution management rules
│   ├── llm_chains.py       # Mistral Cloud synthesis mapping connections
│   └── rag_pipeline.py     # ChromaDB vector indexing and retrieval functions
├── utils/                  # Secondary functional operational aids
│   ├── __init__.py
│   └── exporter.py         # Automated Plaintext and PDF document report builders
├── app.py                  # Streamlit Interface layout UI
├── requirements.txt        # Frozen software package tracking dependencies
└── README.md               # Operations manual documentation asset
🔒 Security & Data Privacy
Zero Cloud Leakage on Speech: Transcription tasks execute locally on your machine via standard CPU allocations. No third-party system holds copies of the raw audio waves.

Credential Integrity: Private API tokens are handled via dynamic runtime memory calls wrapped safely within isolated local .env storage frameworks.