🎙️ YouTube Transcript Chatbot A Python-based chatbot that fetches transcripts from YouTube videos, embeds them using Hugging Face models, stores them in ChromaDB, and allows users to ask context-aware questions about the content. Powered by LangChain, transformers, and Azure AI for LLM-based answering.

🚀 Features 🎥 Transcript Extraction: Automatically extracts English transcripts from YouTube videos.

🧠 Embedding with Transformers: Converts transcript chunks into vector embeddings using sentence-transformers/all-MiniLM-L6-v2.

🧬 ChromaDB Integration: Stores transcript embeddings in an in-memory ChromaDB for fast similarity search.

🗣️ Conversational Q&A: Ask natural language questions about the video content.

🌐 FastAPI Backend: Offers RESTful endpoints for frontend integration or other services.

☁️ Azure OpenAI Integration: Uses Azure-hosted GPT-4o model to answer questions based on transcript context.

🖼️ Sample Interface

🛠️ Project Structure bash Copy Edit 📁 your-project/ ├── chatbot_transcript.py # Transcript processing and embedding chain ├── chatbot_query.py # Querying and conversation logic ├── api_server.py # FastAPI app to expose endpoints ├── .env # Stores Hugging Face and GitHub credentials └── requirements.txt # All Python dependencies 🧰 Requirements Python 3.8+

Hugging Face Transformers

LangChain

ChromaDB

YouTube Transcript API

Azure AI SDK

FastAPI

🔐 Environment Variables Create a .env file in the root directory with the following:

ini Copy Edit HUGGINGFACE_API_KEY=your_huggingface_api_key GITHUB_TOKEN=your_github_access_token_for_azure_models 📦 Installation bash Copy Edit

Clone the repo

git clone https://github.com/your-username/yt-transcript-chatbot.git cd yt-transcript-chatbot

Set up virtual environment

python -m venv venv source venv/bin/activate # or venv\Scripts\activate on Windows

Install dependencies

pip install -r requirements.txt 🧪 Running the CLI Version Run the terminal chatbot interface:

bash Copy Edit python chatbot_query.py Paste a YouTube URL when prompted.

Then ask questions like:

"What is the video about?"

"Summarize the first section."

"What are the main points discussed?"

Type exit to end the session.

🌐 Running the API Server To launch the FastAPI backend:

bash Copy Edit uvicorn api_server:app --reload Endpoints POST /upload-video-url Uploads and processes a YouTube video.

Request Body:

json Copy Edit { "url": "https://www.youtube.com/watch?v=abc123xyz" } POST /query Ask a question based on the uploaded transcript.

Request Body:

json Copy Edit { "query": "What does the speaker say about machine learning?" } GET /response Returns the latest model response.

🧠 How It Works Extract Video ID – From a full YouTube URL.

Fetch Transcript – Using youtube_transcript_api.

Split Transcript – Into overlapping chunks using a custom recursive splitter.

Embed Text – Convert text chunks into embeddings using Hugging Face Transformers.

Store in ChromaDB – Enables fast vector search for later queries.

Query Handling – User queries are embedded and matched to similar chunks.

LLM Answering – GPT-4o (via Azure) responds using the retrieved transcript chunks.

📈 Use Cases Educational video summarization

Customer service video analysis

Podcast and lecture question answering

Content review for accessibility

📋 Future Improvements Persistent storage (ChromaDB with file-backed DB)

Multi-language transcript support

Frontend UI integration (React/Next.js)

Upload support for custom audio/video files