ThoughtScope AI

:mag: Visualizing LLM Document Processing with Streamlit

Hey Streamlit community!

I built something that I think you might find interesting - a Streamlit app that visualizes how LLMs process documents in real-time.

What it does

The app shows you exactly which parts of a document an LLM focuses on when answering questions. It highlights the relevant chunks directly on the page as the model processes them.

Tech Stack

  • Streamlit for the frontend
  • unstructured.io for document processing
  • RAG (Retrieval-Augmented Generation) for document Q&A

Try it out

Currently using MLK’s Letter from Birmingham Jail as a demo document.
Check out the demo

Coming Soon

  • Custom document upload functionality
  • Support for multiple document formats
  • Enhanced visualization features

Looking for feedback

  • How could this be more useful for your Streamlit projects?
  • Any suggestions for UI/UX improvements?
  • What features would you like to see added?
  • Any tips for optimizing performance?

I’m also attaching some cool :sunglasses: screenshot of app that how annotated chunks looks like which evidence score.

2 Likes