AAAI-27 Demonstration Program
Project page / GUI · Video demo · Source code · Instructional Agents
EduCAST turns a topic, a learner profile, and a learning goal into a narrated lesson with animations, illustrations, and interactive practice. An instructional planner writes a learner-conditioned lesson plan, a visual designer turns each scene into animation code or an image prompt, a rendering engine produces narrated media, and a VLM reviewer checks every scene and repairs only the ones that fail.
| Path | What it is | Published at |
|---|---|---|
docs/index.html |
Project page | https://darl-genai.github.io/EduCAST/ |
docs/demo/ |
Static build of the lesson composer and showcase | https://darl-genai.github.io/EduCAST/demo/ |
src/ |
Source code: the EduHarness system, the legacy pipeline, baselines, notebooks |
GitHub Pages serves docs/ from the main branch. The demo is generated; rebuild it after
changing the website under src/eduharness/eduharness/studio/:
cd src/eduharness
python scripts/build_pages.py # writes ../../docs/demo/The online demo cannot generate lessons, because that needs the Python server and local renderers. To run it yourself:
cd src/eduharness
python -m venv .venv && source .venv/bin/activate
python -m pip install -r requirements.txt
python -m eduharness.studio.public --host 127.0.0.1 --port 8091Then open http://127.0.0.1:8091/. Full setup, including the renderers (FFmpeg, Manim,
Playwright, Remotion) and the bring-your-own-key API, is in
src/eduharness/README.md.
- Instructional Agents (EACL 2026): multi-agent course generation, from learning objectives and syllabus to slides, scripts, and assessments. Project page
- Instructional Agents Studio: the course-building app for instructors,
npx instructional-agents.
@inproceedings{chen2027educast,
title = {EduCAST: A Multi-Agent System for Personalized Educational Video Generation with Closed-Loop Review},
author = {Chen, Qingyue and Nandra, Aarshdeeep Singh and Yao, Huaiyuan and Shourie, Varun and Kazi, Rayan and Wei, Hua},
booktitle = {Proceedings of the AAAI Conference on Artificial Intelligence (Demonstration Track)},
year = {2027}
}