CornelliusYW/Multimodal-RAG-Implementation ? reverse-engineered prompt

Build me a multimodal RAG notebook that can take a PDF, pull out the pages as images, process audio files, and let me ask questions across both the images and the audio transcript. I want it to create embeddings for the images and text, save them in a vector store, then retrieve the most relevant pieces when I type a query. After that, use the retrieved images and text to generate a helpful answer in plain language and show the matching images too.

Please make it work as a clear end to end notebook that I can run step by step, with the main pipeline easy to follow and enough comments so I understand what each part is doing. If you need to check current docs for any model details or setup steps, go ahead and look them up online.