swapwarick/LightRAG-OCR — reverse-engineered prompt

Build me a local web app for private document search. I want to drag and drop scanned files like PDFs, images, TIFFs, and text files, have the app read the text on my machine, break it into searchable chunks, and save everything locally so my documents never go to the cloud.

The dashboard should feel polished and simple. Show whether the server is online, stream live processing logs while a file is being imported, support light and dark mode, and show basic stats like how many documents and chunks are indexed. I also want a document list where I can remove uploaded documents and their saved search data.

Add a search box where I can ask a normal language question and get the most relevant text chunks back with similarity scores so I can see why each result matched. Use local OCR, local embeddings, and a local vector database. GPU acceleration is nice if available, but it should still work on CPU.