agiagoulas/page-stream-segmentation ? reverse-engineered prompt

Build me a Python service for page stream segmentation that can take a PDF upload and predict where one document ends and the next begins.

I want it to support text only, image only, and combined text plus image processing, with an option to use either a model based on the current page only or one that also looks at the previous page. Please expose simple API routes for each mode, and make sure the PDF is sent as form data with the key file. It should also include clear Swagger style docs so I can try the endpoints easily in a browser.

If it helps, use the same setup the README describes, with a FastAPI style app, a way to build document streams from PDFs, and notebooks for model training. The project should feel ready to run locally after installing dependencies and placing the needed word vectors in the right folder. If you need current library details, look them up online.