ConardLi/easy-dataset ? reverse-engineered prompt

Build me a web app for creating datasets for LLM fine tuning, RAG, and model evaluation. I want to be able to upload documents like PDF, Word files, markdown, text, and EPUB, then have the app clean the content, split it into useful chunks, and help generate question and answer pairs from it. It should also support multi turn chat style datasets, image based QA, and evaluation sets with different question types like multiple choice, short answer, and open ended. Please include a simple, modern interface where I can review, edit, tag, and export datasets in common formats. It should work with OpenAI compatible model APIs, plus local models if possible, and let me track background jobs, token usage, and generation progress. If you need to, look up current docs online for any setup details.