angelotc/pdf2md ? reverse-engineered prompt

Build me a Python tool that takes a folder of academic PDFs and turns each one into a clean markdown summary. I want it to find the paper title as well as it can, pull out the full text, spot figures and captions, and use an LLM to write a structured summary with things like a quick takeaway, methods, results, and practical notes. It should save one markdown file per paper, plus a simple index file that links everything together, and optionally a combined summary file too.

Please make it work with OpenAI compatible APIs, including OpenRouter and Gemini style endpoints, and let me configure the prompts in a JSON file. It should skip work when nothing changed, so reruns are fast, and it should have a command line script I can run on a single PDF or an entire papers folder. If you need to, look up current docs online for the PDF and vision parts so the extraction is reliable.