Embed the tool in the subject's workflow and design feed-forward prompts, symbolic-answer handling, and progress tracking before full release¶

Strategy · All strategies
Evidence · no claims cited

Description¶

The authors distil six transferable recommendations for educators adopting the approach: "Keep the human in the loop at solution verification, not real-time grading"; design feedback prompts explicitly around feed-forward, locating the error and guiding next steps rather than revealing the answer; invest early in symbolic-answer entry, preview, and algebraic-equivalence handling and pilot with a small group; provide clear progress tracking and topic or question-type selection; embed the tool in the subject's workflow, for example by linking it to specific topics or low-stakes checkpoints, rather than releasing it as a standalone optional resource; and be transparent that the AI can hallucinate while solutions are human-verified.

Design Implications¶

Context¶

Requirements¶

• Feed-forward prompt design is described as the hardest behaviour to deliver consistently
• Symbolic and notation-heavy answers need careful input handling

Constraints¶

• Voluntary, unweighted activities attract uneven engagement; generating feedback for every submission carries time and compute costs students notice

Target Learners¶

• engineering educators teaching large mathematical subjects

Target Learning Goals¶

• trustworthy scalable formative practice
• student-directed practice and calibrated trust in AI feedback

Related Strategies¶

• Treat AI as a guided support tool and explicitly build verification and skepticism into learning goals

Examples¶

-

Key Sources¶

• Lili Chen, Gavin Buskes, Yuxin Ren and Chin Tong Leong. (2026). Feedback Without the Wait: Piloting a Generative AI Practice Platform in a Large Maths Class. 37th Annual Conference of the Australasian Association for Engineering Education, UNSW, Sydney. https://creativecommons.org/licenses/by-nc/4.0/