deepsearch-ai/deepsearch ? reverse-engineered prompt

Build me a Python app for semantic search over multimedia files like images, audio, and video.

I want to be able to upload local files or point it at an S3 path, have the app use models to describe and tag the content, turn that into embeddings, and store it in a vector database so I can search it later with natural language. The search results should not just return the matching file, but also show the most relevant part, like the exact second in an audio file or the most important moment in a video.

It should have a simple Streamlit style web UI for browsing and searching my media, and a small Python interface too so I can add data and run searches from code. If it helps, look up the current docs for the libraries you need and use sensible defaults for the models and database, but keep it easy to configure. Also include support for YouTube channel video ingestion and a clear place to set the required API key.