Hailiang Launches HaiMate, a Teacher AI Agent Built on Its Own 150 Schools

Hailiang Technology, the education technology arm of Hailiang Education Services, used the 88th China Education Equipment Exhibition in Changsha on 30 October to launch HaiMate, an AI workbench built specifically for teachers. The product folds AI lesson preparation, classroom review, grade analysis and teacher profiling into a single conversational entry point: a teacher states a need and moves through lesson design, courseware generation, learning diagnosis and management reporting in one loop.

"Teachers don’t lack a chat-capable AI; they lack a work partner that really understands schools," HaiMate product manager Shen Yunqi said at the launch. Hailiang itself runs more than 150 schools with over 200,000 students and teachers, and Shen argued HaiMate "grew out of lesson-prep books and teaching-research meetings" rather than from a generic model; the company draws on more than three decades of school operation.

In lesson preparation, a teacher selects stage, subject, lesson type and style, then generates a three-piece set — learning guide, lesson plan and teaching slides — in one click, with local "AI rewrite" for passages that don’t work before export. Preparation that used to involve collecting material and fixing formatting now takes about half an hour, the company said.

Shen demonstrated the difference against a general-purpose agent using Grade 7 English, Unit 3 "Keep fit", Section B. The general agent produced a list of bullet points with no reference to the textbook and even included out-of-scope content; HaiMate first extracts the textbook’s knowledge points so the output hangs on the textbook knowledge system with visible chapter mapping, and arranges lesson plans according to teaching paradigms drawn from education specialists.

After class, HaiMate generates a review report from a lesson recording or transcript, scoring teaching quality, method and student interaction and cross-mapping Bloom’s cognitive levels against classroom interaction to distinguish genuine engagement from surface response, rather than settling for "good classroom atmosphere". On grades, importing several exam cycles, selecting a batch and clicking generate produces trend lines, class and subject comparisons and identified weak points with charts and commentary; managers can then keep asking follow-up questions in the chat and drill into the same dataset.

For school leadership, HaiMate links teacher profiles, performance data, awards and lesson-observation records into a "teacher nine-box grid" that segments talent along two dimensions — a principal can ask, for example, for three candidates suited to a final-year homeroom role and receive recommendations with the reasoning and evidence attached. HaiMate’s stated boundary is that repetitive, standardisable work goes to AI while judgements about learning, instruction and student development stay with the teacher. As Shen put it: AI cannot take the podium for a teacher, but it can save them a few all-nighters.