This version introduces a high-performance Agentic RAG pipeline that enables intelligent, real-time conversations with YouTube video transcripts. By leveraging Qwen2.5-7B and a two-stage retrieval process with Cross-Encoder reranking, the system provides highly precise, hallucination-free answers accompanied by clickable timestamp citations. Key technical enhancements include on-the-fly multilingual translation, semantic overlap chunking to preserve context, and a map-reduce sampling strategy for summarizing multi-hour videos.

Initial stable release of YT RAG Assistant. Features include YouTube transcript extraction, syncing to a Hugging Face backend, and an AI chat interface to query video content.