Mestria Research
We’re investigating how AI can help learners build understanding, recognise misconceptions and apply what they know independently.
Our starting point is a teaching question. The technology earns its place by helping us answer it.
Our measure of progress
Better teaching.
More independent thinking.
The research agenda
Current methods provide the starting point. Each research direction describes the next question we want to investigate.
Does the assistant help learners work through the next step?
In use today
Our teaching rules guide the assistant toward hints, questions and explanations. Scripted behaviour checks test whether it coaches a learner or gives away an answer.
Planned research
Test a wider range of simulated learning sessions, with different misunderstandings and starting points, to improve how the assistant guides each next step.
What makes a problem challenging for a learner?
In use today
Practice content goes through answer verification and separate difficulty reviews. Correctness and difficulty are treated as different questions.
Planned research
Improve difficulty reviews by examining the reasoning a problem requires and using carefully reviewed examples as reference points.
Which missing foundation would make the next step possible?
In use today
Concept lessons, guided practice and review feedback give learners ways to revisit an idea and examine an approach.
Planned research
Ask subject experts to review which concepts a learner needs first, and investigate recurring misunderstandings so guidance can address the underlying gap.
Can learners apply what they know later, without assistance?
In use today
Practice and feedback help learners examine their reasoning and see where they need more work.
Planned research
Study how practice progress relates to what learners can do independently later, what they remember and how they apply it to new challenges. Examine both how well we measure understanding and how teaching helps it grow.
Inside the work
Notes on what we check today and how we intend to test what comes next.
Current method
Method note ·
Our SAT and ACT practice questions are checked by three independent AI solvers, with answers settled by majority consensus. Difficulty is reviewed separately.
These AI checks help establish an answer key; they do not constitute human approval of every item. Our research explores how difficulty checks can better reflect the way people learn.
Independent research also informs our approach: Bastani et al., PNAS (2025) examines how the design of AI guidance affects later unaided performance in high-school mathematics.
Research direction
Research note ·
We’re developing simulated learning sessions to explore how the assistant responds to misunderstandings, guesses and requests for an answer.
We want to check whether its guidance helps a learner take the next step while leaving the thinking in their hands. Studies with real learners will help us understand what they learn and retain.
How we move forward
We intend to publish findings with their methods, scope and limitations, so learners, educators and families can understand what was measured.
Our research plans call for subject experts to review proposed connections between concepts, and for recommendations to come with reasons a learner can understand.
We use simulated conversations for tutor-training research, not minors’ conversation logs.
See our approach to learning