RAGit!

A framework to create a RAG assistant for videos.

To Do

• Get text from video
  • Youtube API (unofficial)
  • GCP StT automatic
  • Local StT models (abstract class)
• Function to perform embedding (multilingual gecko)
• Lift Vector DB (vector search? or Chroma?)
• RAG prompt (query answer)
• Generate citations to document/chunk (Check grounding? Citation?)
• Get starting time for a chunk
• Wrap up all together

Extra mile

• Implement Semantic Chunking to create chunks (default is len words)
• Optimize StT by adding punctuation (LLM calling)
• Dynamically select the prompt by using the language in the video (worth it? what if query is in a != language?)
• Enable RAG fusion

Videos

1 - El ORIGEN de los distintos ACENTOS de Argentina, id: NgbEL2HbXWw