Skip to content
3 sourcesPDF, video, and web ingestion behind one path
End-to-endfrom ingestion to action dispatch
Extensibletool and reasoning layers

The problem

Most conversational demos stop at question answering. Real workflows need to ingest messy sources, keep context across turns, and dispatch a concrete action.

Approach

  • Built extraction across PDFs, video, and web URLs behind a single ingestion path.
  • Kept the agent layer extensible so reasoning steps and tools can be added without rewriting the workflow.
  • Connected the reasoning stage to an action-dispatch path rather than ending at a text answer.

Stack

  • Python
  • LangChain
  • Groq
  • Gradio
  • EasyOCR

What these numbers do and do not prove

This is an open-source framework rather than a deployed product. Judge it on the repository design and documentation.

Source

Other work

Start a conversation