https://github.com/SulemanKazi/TtYT

I built this little application for setting up RAG on Youtube video transcripts. I haven't seen a lot of RAG for multimodal use-cases so I thought this was interesting. It uses YoutubeDL to get transcripts and Vectara to set up the RAG system. It's multilingual so other languages besides english work as well.

The answer produced by the LLM has references with timestamps to where its deriving information from from within the video. Right now the interface is a clunky terminal but I'm planning on adding a UI or maybe a voice I/O interface to the app.

Full disclosure: I work at Vectara, but this was a personal/side-project!