05 · Project
Personalized AI Teaching Copilot
A personalized AI assistant for teachers and students in Bijie, Guizhou.
- 543 textbooks in 14 subjects
- 95,360 searchable curriculum chunks (the dataset is private)
- <30 s class-prep draft (median 18.9 s) instead of about an hour
- ~150 students used it (project lead’s count)
Built with
- RAG
- ChromaDB
- Qwen
- Next.js
- Express
- Mermaid
About the project
For teachers: a RAG assistant (retrieval-augmented generation: it looks things up in the textbooks before it writes) built on ChromaDB and Qwen over 543 textbooks in 14 subjects. It gives a class-prep draft in under 30 seconds (measured median 18.9 s) instead of about an hour. 10+ teachers used it.
For students: a Next.js and Express app with leveled quizzes and Mermaid mind maps, used by about 150 students.
In Bijie, about 100 seventh graders in two classes took part in a class-level comparison, one class per condition and not randomized. The overall test-score difference was not statistically significant, so the project does not claim a learning-effect improvement.
Worth knowing. A class-level comparison of two classes of about 50 seventh graders, not randomized: the overall test-score difference was not statistically significant, so the project does not claim a learning-effect improvement.
The public demo uses synthetic data.
In the field, weak internet and server costs were real limits.
Next project