Socratic AI Tutor

Situation

Students use AI to finish assignments and learn nothing from them. ilm's bet was a tutor that refuses to answer — but that bet only pays if the refusal survives a real student pushing for the answer.

Task

• Prove the Socratic line holds under pressure, not just in a demo.
• Show it working by voice, since that is how tutoring actually sounds.

Result

A working proof of concept that held the line — guiding questions, by speech or text — and helped win ilm a $12,000 WA Government Innovation Grant.

My Contribution

• Authored the technical proposal and requirement artifacts the grant application rested on. The proof of concept is what let ilm demonstrate the idea in the room rather than describe it.
• Put the constraint in the prompt rather than hoping for it: an answer only after two or three hints, and only for general knowledge. 'Be Socratic' does not survive a student who keeps asking.
• Aimed the product at the policy question, not just the classroom one. Australia's concern is children using AI safely, and a tutor that refuses to do the homework is an answer to it.
• Chose the browser's own speech engines, so voice costs nothing per minute — the difference between a pilot a school can afford and one it cannot.