YadneshBamne/pb4-prototype ? reverse-engineered prompt

Build me a working demo of Token Diet, a smart context compression tool for RAG.

I want a simple app where I can paste a user question and some retrieved text, then see how the system compresses that context before sending it to an LLM. It should show the full pipeline clearly, including retrieval style scoring, redundancy removal, cleanup of filler language, token budget packing, and a before versus after comparison with the final compressed context.

Make the interface easy to understand, with a clean frontend and a Python backend that can process the text locally. I want the app to highlight which sentences were kept, which were removed, and why. If it helps, use local document search and modern NLP models, and look up current docs online if you need to. Keep it as a prototype that makes the idea visible and interactive, not a huge production system.