🎥 VidIntel

An AI-powered YouTube video summarization and question-answering app built with Gradio, LangChain, FAISS, and IBM watsonx.ai. This application extracts YouTube transcripts, summarizes long videos, and allows users to ask contextual questions grounded strictly in the video content.

---

🚀 Features

• 📜 Automatic YouTube Transcript Extraction
• ✂️ Smart Text Chunking using LangChain splitters
• 🧠 Semantic Search with FAISS
• 🧾 Concise Video Summarization
• ❓ Context-aware Question Answering
• 🔗 IBM Granite LLM (watsonx.ai)
• 🌐 Gradio Web UI

---

🧠 Architecture Overview

YouTube URL
   ↓
Transcript Extraction (youtube-transcript-api)
   ↓
Preprocessing & Chunking (LangChain)
   ↓
Embeddings (IBM Slate-30M)
   ↓
FAISS Vector Store
   ↓
LLM (IBM Granite 3.2)
   ↓
Summary / Q&A Output

---

🛠️ Tech Stack

Layer	Technology
UI	Gradio
LLM	IBM Granite 3.2 (watsonx.ai)
Embeddings	IBM Slate-30M
Framework	LangChain
Vector DB	FAISS
Transcript	youtube-transcript-api
Language	Python 3.10+

---

📦 Installation

1️⃣ Clone the Repository

git clone https://github.com/PRONGS-CHIRAG/Vidintel.git
cd Vidintel

2️⃣ Create Virtual Environment

python -m venv venv
source venv/bin/activate  # macOS/Linux

3️⃣ Install Dependencies

pip install -r requirements.txt

---

🔑 IBM watsonx.ai Setup

You must have access to IBM watsonx.ai.

Required:

• IBM Cloud Account
• watsonx.ai project
• Project ID

The app currently uses:

project_id = "skills-network"
url = "https://us-south.ml.cloud.ibm.com"

⚠️ For production use, store credentials securely using environment variables.

---

▶️ Run the Application

python app.py

Then open:

http://localhost:7860

---

🧪 How to Use

🔹 Summarize a Video

1. Paste a YouTube video URL
2. Click “Summarize Video”
3. Receive a concise AI-generated summary

🔹 Ask Questions

1. Enter a question about the video
2. Click “Ask a Question”
3. Get an answer grounded in the transcript

---

🧩 Project Structure

.
├── app.py                  # Main Gradio application
├── requirements.txt        # Python dependencies
├── README.md               # Project documentation

---

🧠 Models Used

LLM

• ibm/granite-3-2-8b-instruct

Embeddings

• ibm/slate-30m-english-rtrvr-v2

---

⚠️ Known Limitations

• ❌ Works only with videos that have English transcripts
• ❌ Private / restricted videos are not supported
• ⚠️ Transcript fetching depends on YouTube availability
• ⚠️ Cold start latency due to LLM initialization

---

🔮 Future Improvements

• ⏱️ Timestamp-aware answers
• 🌍 Multi-language support
• 💾 Persistent vector storage
• 🔄 Streaming responses
• 🔐 Secure credential handling
• 🚀 Deployment on Hugging Face Spaces

---

🤝 Contributing

Contributions are welcome!

1. Fork the repository
2. Create a feature branch
3. Commit your changes
4. Open a Pull Request

---

📄 License

MIT License Feel free to use, modify, and distribute.

---

🙌 Acknowledgements

• IBM watsonx.ai
• LangChain
• Gradio
• FAISS
• YouTube Transcript API

---

📬 Contact

If you’re building AI, RAG, or agentic systems and want to collaborate:

Chirag 🔗 LinkedIn - https://www.linkedin.com/in/chiragnvijay/ 💡 Open to research, startups, and applied AI projects

---