Begin with a teaching problem
Changjiang Daily reported in February 2026 that Wuhan Duanyu Intelligence founder Qin used his English-teaching experience and AI to develop Phrase AI around phrase learning and teachers' tasks. This is a research starting point, not a judgment of learning effectiveness.
Editorial analysis: software completion and teaching effectiveness need separate acceptance
At least two questions matter: whether the system records learning actions correctly and whether exercises fit target learners. AI can assist code and materials, but development speed does not prove educational effectiveness. Agree learners, pacing and feedback, recording technical faults separately from teaching judgments.
A trial could have teachers select a few phrases, check meanings, examples and context, then observe learners completing tasks. Record misunderstandings and system responses without unrelated data collection. Clearly describing tested scope and uncertainties is more informative than calling one demo universally effective.
Let experience produce a question list
Reported tools and progress are historical. Current schools served, payments and staffing are unverified here. Further reading: accepting a sample lesson.