CardTutor

Inspiration

Learning with LLMs often feels slower than it should:

• Answers are verbose when you want a quick explanation.
• You get hit with unfamiliar terms without any scaffolding.
• Linear chat makes you scroll, lose context, and never really track what you actually learned.

We wanted something that feels like Canvas, but built for AI-first learning.

What it does

TutorCanvas lets you create Courses → Modules → Items, where an “AI tutor thread” is not a wall of text but structured learning.

Each tutor response is converted into:

• Core answer (short and direct)
• Expandable cards (Intuition / Example / Formal / Pitfalls)
• Clickable term tooltips + a course Glossary drawer
• A micro-check question to confirm understanding

Progress is tracked by checkpoint performance + term mastery, not by time spent.

How we built it

• Frontend: Canvas-style layout (course sidebar, modules list, main content, glossary drawer)
• Backend: API routes for courses/modules/items, chat history, and progress computation
• AI layer: the LLM returns strict JSON so the UI can reliably render cards, terms, and micro-checks

We store:

• tutor thread events (questions + structured answers)
• glossary terms with statuses (unknown / learning / known)
• micro-check attempts for progress

Challenges

• Structured output reliability: we enforced a strict JSON schema and validated responses to prevent UI breakage.
• Making it feel non-linear: we avoided “scrollback hell” by using cards + a glossary drawer instead of pure chat.
• Progress that means something: we tied progress to demonstrated understanding (micro-checks + term mastery).

What we learned

LLMs become much better tutors when you control:

• pacing (short core + expand)
• vocabulary (term tooltips + glossary)
• feedback (micro-checks)
• structure (course/module organization)

Next steps

• Import PDFs/lecture notes and auto-generate modules + tutor threads
• Spaced repetition from “unknown” terms and missed micro-checks
• Better personalization (user skill model per course/topic)

Built With

• next
• prisma
• tailwind