AI product Open source

Whiteboard AI Tutor

Whiteboard AI Tutor is an open-source AI tutoring reference implementation and LLM benchmark. An LLM streams JSON drawing operations—writing, lines, rectangles, ellipses, paths, and erasing—to a browser player, which renders them as handwriting on a 1600×900 board while accompanying narration is displayed. When a student interrupts, the application stops the stream and sends the question together with the current board state back to the model, which can answer and annotate or erase existing objects by ID.

View repository Visit site Mentioned in 1 video ↓

Overview

The repository includes a player, protocol, model runners for Claude, Gemini, and DeepSeek, eight fixed science, mathematics, and economics scenarios, and recorded lessons that can be replayed in a browser. It also contains a benchmark using one frozen prompt, six models, two thinking settings, and repeated runs, with automatic protocol and latency metrics, linting, rendered boards, and blind review. It is not a hosted tutoring service: recorded lessons can be replayed without API keys, while generating new lessons requires running the benchmark tools with model API keys. The repository is licensed under MIT.

What Whiteboard AI Tutor is used for

1 use taken from transcripts — each links to the moment in the video.

  • An open-source tutor that teaches on a whiteboard: a model streams JSON drawing instructions that a browser player turns into handwriting, interruptible mid-lesson; the repo also benchmarks six models on 288 blind-reviewed lessons.

Videos mentioning Whiteboard AI Tutor

1 in the library.