📝 Phase 2: Transcript Extraction & Vector DB Setup

## Phase 2: RAG Foundation

Implement automated transcript extraction and vector storage.

### Tasks
- [ ] Implement Transcript Downloader service using yt-dlp
- [ ] Set up Vector DB connection (Redis/Chroma)
- [ ] Create Embedding pipeline (Text → Vector)
- [ ] Design chunking strategy for long transcripts
- [ ] Implement batch processing for multiple videos
- [ ] Add error handling for videos without captions

### Acceptance Criteria
- Automatic transcript extraction for all course videos
- Transcripts chunked optimally (512-1024 tokens)
- Embeddings stored in vector database
- Fallback mechanism for caption-less videos
- Metadata stored alongside vectors (video_id, timestamp)

### Tech Stack
- yt-dlp for captions
- Redis Vector Search / ChromaDB
- sentence-transformers or OpenAI embeddings
- Langchain for chunking