Confucius, Interactive AI Tutor

Interactive AI tutor that turns your lectures into 90-seconds crash courses

Comment

🎓 About the Project

Confucius: Your Interruptible AI Tutor

🧠 Inspiration

Long educational videos are valuable but slow. Students rarely rewatch entire lectures, and traditional summarizers lose context or nuance. We wanted something personal, interactive, and fast.

We asked ourselves:

“What if learning felt like having a wise tutor beside you one who pauses the video whenever you ask, explains the hard part, and condenses everything into a 90-second lesson?”

That idea became Confucius: an AI tutor that sits next to your video, answers your questions in real time, and distills a 15-minute lecture into a crisp, 90-second crash course narrated by a talking Confucius avatar.

🌟 What Confucius Does

Confucius is a full AI-powered learning assistant with:

⭐ 1. 90-Second Crash Courses

Automatically transforms long educational videos into structured, accurate, digestible summaries.

⭐ 2. Interruptible Tutor Mode

You press Ask Confucius, and:

  • the video pauses
  • Confucius explains the concept with examples, analogies, and diagrams
    Just like a real tutor sitting next to you.

🚀 How We Built It (Technical Overview)

1. Transcript Extraction

We use Whisper and OCR to generate segmented, time-aligned transcripts from uploaded lecture videos.

2. Semantic Chunking

We break the transcript into small, meaningful parts using embeddings and natural pauses or topic changes. This produces a clear structure.

3. Slide Generation

We convert each chunk into a short slide with:

  • a title
  • a few key points
  • quick speaker notes

This forms the 90-second crash course.

4. Tutor Mode

When the user clicks Ask Confucius:

  • we pause the video
  • locate the closest transcript section
  • send it to the LLM for a simplified explanation

The user gets an instant mini-lesson.

5. Pipeline Coordination

A single backend script connects everything:
video → transcript → chunks → slides → explanations
so the frontend only needs to send one request.

Built With

Share this project:

Updates