para-lost/RVP ? reverse-engineered prompt

Build me a working setup for this Recursive Visual Programming project so I can run it on my own images and videos. I want the app to take a CSV with a question, an optional answer, and an image or video file name, then run the recursive visual reasoning pipeline and save the results in a way that is easy to review and score.

Please make sure the project can be launched from one simple command after setup, and that the configs, model paths, and API key are handled cleanly. If anything is missing or needs current instructions, look up the latest docs online and make the setup as smooth as possible.

I’d also like the code to support the newer vision models mentioned in the repo, and to work for both recursive and non recursive mode. If there are obvious rough edges in the current scripts or setup steps, clean them up so a non expert can follow along without getting stuck.