Photograph a page of student work. 拍懂 turns the errors into a likely learning bottleneck a teacher can review, and a useful next teaching move — then keeps a record of what has been checked since. Key judgements are always reviewed by a teacher.
A mark out of ten does not say where the work broke down. Photograph the page and 拍懂 turns the errors into a clear, likely bottleneck — a concept, the reading of the question, units, or a step skipped too quickly. Common handwritten answers can be read by the AI first; anything it is unsure of goes to a teacher for review.
“One place value missed. Next: three short follow-up questions with Murph.”
Not ten more questions — the one thing worth teaching next. 拍懂 puts the possible stumbling points in order and offers one short, specific path. Key judgements go to a teacher for review — which is the real reason families are willing to trust what comes back.
Finishing the practice is not the point; the check is. Progress sits on a nine-square grid — each concept worked through lights one square, and a full row means the whole line has been checked, not just one question. A quick check the next day (T0.5), then a follow-up check a few days later (T1).
Source: operating records from xylex’s own teaching site (test rows excluded; usable records only). This page makes no claim about a specific learning-gain percentage. Every product screen shown here is our own mock-up and contains no copyright-protected exam material.
Student data is used for diagnosis and teacher review only. Real names and original answer images are not shown publicly; anything shown externally — including every product screen on this page — is de-identified or a mock-up we made ourselves. Every marking record is kept in full and can be traced back, and key judgements can be reviewed by a teacher.
拍懂 (STEM Insight) is an AI learning-diagnosis product built by 浩群數位科技有限公司, a company registered in Taiwan. xylex is the cognitive-diagnosis and human-review engine behind it. The same engine supports three applications.
For teachers and families: photograph the work, surface the likely bottleneck, have a teacher review it, and watch progress as each step is checked. Students who are currently struggling get more precise support without being labelled; students who are ready to progress are not held back by another ten questions they can already do.
Diagnosis, anti-hallucination checks and teacher review offered as infrastructure, so other AI teaching products can connect to them directly.
The same engine extended to skills diagnosis and outcome checks for workplace training and technician certification.
Fifteen minutes is enough to see how 拍懂 turns “they just don’t get it” into something you can diagnose, teach into and check. Or request pilot access and run it on your own students’ work.