PipThePilot
An AI that teaches the four forces of flight has to be stopped from just answering the question, or it isn't teaching anything at all.
- Human-in-the-loop routing
- Socratic prompting
- Instructional design
The classroom problem
Bernoulli’s principle and the four forces of flight are exactly the kind of topic where a direct answer feels like it helps and actually doesn’t. A student who’s told “lift happens because air moves faster over the curved top of the wing” can repeat that sentence on a quiz without having built any working model of why it’s true. The forces-of-flight unit is small enough to fit in a single lesson and rich enough that most of the value is in the reasoning a student does to get there — which is precisely the part a direct-answer tool skips.
The design decision
PipThePilot never gives a direct answer. That’s not a style choice, it’s the entire pedagogical argument: an AI tutor that answers a physics question outright has taught a student to ask an AI physics questions, not to reason about physics. The system is designed as human-in-the-loop by necessity, not convenience — the teacher fields every question from students first and routes it to the AI, which then works Socratically, asking the question back in a more answerable form rather than resolving it. The teacher stays the one deciding which questions get routed and when a student needs a human explanation instead.
How it works
The intended flow, as designed for a Masters in Education capstone:
- A student asks a question about lift, drag, thrust, or weight during a lesson.
- The teacher fields it and decides whether to route it to Pip.
- Pip responds only with guiding questions — never a direct statement of the answer — walking the student toward Bernoulli’s principle themselves.
- The teacher can end the exchange and step back in at any point.
What I’d do next
PipThePilot exists right now as a presentation and a set of screenshots from its Masters capstone — not a running application. That’s worth saying plainly rather than implying a working demo exists: a well-argued design case study is more honest, and more useful to a hiring committee, than a half-built prototype that undersells the idea behind it.
The natural next step is a scoped-down build: a single-topic Socratic tutor, gated entirely behind a teacher-operated interface, tested with one class on one unit before deciding whether the human-in-the-loop routing model holds up outside a slide deck.