fluffy-guide

A real-time lecture copilot that maintains rolling context and lets learners ask quiet questions without interrupting the speaker.

Lecture confusion is time-sensitive. If someone misses a definition or transition, waiting until the end can make the rest of the session harder to follow.

Inside the experience

What someone can actually do

  1. Rolling transcript and short evolving summary
  2. Silent text or voice questions from the learner
  3. Answers grounded in recent lecture context
  4. Demo mode for reliable presentations
  5. Future direction toward slides and XR/spatial input

How the idea took shape

A two-agent Streamlit prototype: one agent captures lecture audio and maintains a rolling summary while another answers questions from recent context.

Lecture confusion is time-sensitive: help is most valuable while the missed context can still repair the rest of the session.
Explore how it was built +

I framed the learning problem and built the transcription, rolling-context, grounded-question, and reliable demo flows.

  • Python
  • Streamlit
  • faster-whisper
  • OpenAI API
  • Realtime context buffers

Where it could go next

The project clarified that a live copilot needs disciplined context boundaries and consent as much as it needs fast transcription.

More work