Shawn Jansepar is Khan Academy’s director of engineering for its learning platform and an engineering and product leader behind Khanmigo, its AI-powered tutor and teaching assistant. He builds educational technology that helps students reason through problems, gives teachers visibility into their progress, and remains practical for real classrooms.
Jansepar studied computing science at Simon Fraser University and interned at Electronic Arts before joining Mobify, a Vancouver mobile-commerce company later acquired by Salesforce. Over six years, he progressed from junior developer to director of engineering. His technical writing explored automated responsive-image optimization and combining native and web interfaces to improve mobile applications.
He also taught at Lighthouse Labs and Simon Fraser University, experiences that helped draw him to Khan Academy. There, he initially managed engineering for independent learners, contributed to a transition from Python 2.7 to Go, and eventually assumed broader responsibility for the learning platform and educational AI.
Designing AI that preserves the work of learning
Jansepar helped shape Khan Academy’s collaboration with OpenAI into Khanmigo, an educational assistant designed around the demands of tutoring instead of unrestricted answer generation. His approach includes several distinctive priorities:
- Socratic tutoring that protects student thinking. Khanmigo uses questions and hints to guide students toward answers, incorporating their coursework, submitted responses, and worked solutions. Mathematical input, graph rendering, multilingual support, and text-to-speech make the experience more useful than a generic chat interface.
- Writing support with teacher visibility. Students receive feedback through outlining, thesis development, drafting, and revision without having the assistant write assignments for them. Teachers can inspect interaction histories and identify unexplained additions to student work.
- Tutoring accuracy as an engineering problem. Effective assistance requires identifying how a student reached an incorrect answer, not simply solving the exercise. Jansepar’s teams combine mathematical agents, calculators, Python, retrieval, multistep prompting, human review, and model-graded evaluations to improve tutoring quality and detect regressions.
- Classroom-ready AI infrastructure. Model routing, reusable platform components, tracing, dedicated inference capacity, and shared-capacity fallback help balance reliability, responsiveness, and cost. His teams prototype rapidly with educators while applying stricter accessibility, testing, and safety standards as features mature.
More recently, Jansepar has led development of a redesigned classroom learning platform created with school-district partners. The initiative combines mastery-based learning, student motivation, accessibility, clearer teacher insight, and alignment with district instructional priorities.
