TownSparrow/S-GAS_Manager ? reverse-engineered prompt

Build me a local web app for trying out an adaptive RAG system for small language models.

I want a simple chat interface where I can send messages, upload documents, and see the system use the uploaded files to improve answers. It should also keep session IDs, show basic session and document info, and have a stats page so I can inspect what happened in a conversation. Please add a benchmark page too, with a way to run and compare different retrieval modes like a basic baseline, hybrid RAG, and the smarter scoring mode.

The app should work on Linux, start from a single entry point, and include scripts for setup and launch. Use a FastAPI backend with a small web frontend, and make sure the retrieval flow includes vector search, graph based context expansion, reranking, and logging for each session. If you need to check current docs for any of the model or server pieces, go ahead and look them up online.