Does Wavesteam have a relevant AI education case study?
Yes. Wavesteam's currently public AI education example is an AI Chinese-learning product, demonstrating conversational language practice, learning-content organization, progress records, and learner feedback. It does not prove experience or outcomes for every AI education category, such as school homework solving, automated grading, teacher planning, or corporate training. Each new scenario needs its own learning objective, learner group, authorized content, and evaluation.
A public case helps a client inspect a real product direction; “AI education” alone combines very different responsibilities.
When applying case experience to a new project, also compare Which industries has Wavesteam worked in? and How can clients verify Wavesteam's project and long-term support experience?; the linked guidance adds context that should be considered in the same decision.
| Product | AI task | Evidence that matters | Relation to the public case |
|---|---|---|---|
| Language practice | Situational conversation, correction, hints, exercise sequencing | Recognition error, response delay, teacher agreement, learning gain | Relatively close in interaction and lesson structure |
| Curriculum Q&A | Retrieve approved material, explain, cite | Citation correctness, refusal, out-of-scope answers, teacher review | Some shared dialogue patterns; new corpus and retrieval evaluation |
| Grading and diagnosis | Score work, identify errors, draft feedback | Agreement with teachers, subgroup bias, appeal overturns | Separate rubric and human-review design required |
| Teacher assistant | Draft lessons, questions, activities, media | Teacher adoption, factual error, editing time, source rights | Different user, content chain, and responsibility |
| Adaptive learning for children | Recommend level and path over time | Age suitability, safety, minimal data, learning gain | Cannot be inferred; child-specific design required |
Success cannot be reduced to smooth conversation or daily activity. A learning-effect claim needs a pre/post design, comparison, or blinded educator review. A teacher-efficiency claim needs time per assignment or lesson, edit rate, and error rework. High satisfaction without measurable educational value does not establish teaching effectiveness.
UNESCO's current guidance on generative AI in education and research calls for a human-centred, age-appropriate approach, data privacy, ethical validation, and pedagogical design rather than treating the technology as a universal answer. UNICEF's 2025 Guidance on AI and Children 3.0 addresses safety, privacy, fairness, transparency, accountability, inclusion, and children's best interests. Those principles translate into parent or educator control, escalation for sensitive topics, minimal retention, age suitability, explanations, and appeal.
Content rights must be established before development. Client-owned teaching material still needs permission for digitization and model processing. Publisher or question-bank content is not automatically available for training because it can be searched. Teacher uploads and learner responses need role-based access, retention, and deletion. A generative model should not become an unreviewed final decision-maker for high-impact grading.
Wavesteam begins with a focused learner range, authorized material, and measurable learning objective. The evaluation covers ordinary items, common misconceptions, out-of-scope and adversarial prompts, and unsafe content, with correct, partially correct, unsupported citation, failure-to-refuse, and educator-edit outcomes. Evaluation samples stay separate from prompt development.
The public case can inform conversation, content, and progress design, while educators co-own the new rubric. Construction proceeds only when users and age are defined, content rights are clear, baseline evidence exists, educator review has an owner, and the proof meets pre-agreed thresholds. Otherwise, the responsible recommendation is to pause rather than launch and repair later.