Knowledge graph
Every AP Physics concept, linked to its prerequisites, the topics that teach it, common misconceptions and the formulas it uses. When a student struggles, the graph shows which earlier idea to repair first.
How it works
A mentor who teaches, and an engine that keeps learning about the student.
APPhysics.si is built around one loop: assess, understand, plan, teach, practise and adapt. Here is what happens at each step and who is responsible for it.
The learning loop
1.Diagnose
A free adaptive diagnostic maps concept mastery, prerequisite gaps and problem-solving habits.
2.Model
The student model records what the student has demonstrated, concept by concept, with how confident we are.
3.Plan
A learning path orders concepts by prerequisites, goals, exam date and time available, and says why.
4.Teach
The mentor teaches 1-to-1, guided by a class brief and free to override any recommendation.
5.Practise
Adaptive daily practice and assignments target the concepts and errors that matter most now.
6.Evaluate
Work is graded with deterministic checks, rubrics and AI reasoning, and uncertain cases go to the mentor.
7.Adapt
Errors and mastery update the student model, and the next activity changes to match.
Inside the Physics Intelligence Engine
Student model
For each concept we track a mastery estimate, how confident we are in it, recent performance, difficulty-adjusted results and when it was last shown. Mastery is an internal learning indicator, not an exact measurement.
Assessment engine
Numerical answers are checked by deterministic calculation with unit and dimension checks. Free-response answers are scored point by point against a rubric, and each decision records the evidence, the error found and a confidence level.
Error intelligence
Mistakes are classified, for example conceptual, algebraic, sign or units, and tracked over time, so a recurring pattern becomes a targeted intervention rather than more of the same practice.
Learning path
Next steps are ordered by prerequisites, current mastery, error patterns, exam date and available time, and every recommendation comes with the reason it was made.
Mentor copilot
Before each class the mentor gets a brief. After it, they can ask for homework on what was taught, a short intervention set, or a summary of the last month. The mentor always has the final say.
Safeguards on automated decisions
- Numerical physics is never decided by a language model alone when it can be calculated.
- Every AI assessment carries a confidence level. Below the review threshold, it goes to the mentor first.
- Mentors can override any automated score, recommendation or misconception flag.
- Student uploads are treated as untrusted input and cannot change how the system behaves.
- Names and contact details are not sent to AI models, and student work is not used for training.
Start with what you already know.
The free diagnostic shows which concepts are secure, which prerequisites are missing and where your mentor should begin.