Youtube Video Transcript Summarizer Using Ai
Ask questions from any YouTube video using AI + embeddings + Svelte frontend.
YouTube Q&A AI App
Ask questions and get concise answers from any YouTube video using embeddings, local language models, and a sleek Svelte frontend.
Features
- Extracts transcripts using YouTubeTranscriptAPI or Whisper (fallback)
- Embeds transcript chunks with
MiniLMviasentence-transformers - Uses FAISS for fast vector similarity search (RAG)
- Answers generated via
Mistral(or any Hugging Face text-gen model) - Persistent vector store with FAISS + local JSON
- Svelte frontend with loading/error states and chat history
Tech Stack
| Layer | Tool/Library |
|---|---|
| Backend | FastAPI, FAISS, Transformers |
| Embeddings | sentence-transformers (MiniLM) |
| LLM | Mistral-7B-Instruct (local) |
| Fallback | Whisper for audio transcription |
| Frontend | Svelte + Tailwind (minimal) |
Getting Started
1. Backend Setup
cd backend
pip install -r requirements.txt
uvicorn main:app --reload
cd frontend
npm install
npm run dev
