Larkspur Learning
A course platform with an AI tutor chatbot that answers student questions against the actual course material, available around the clock.
Industry
Education
Timeline
10 weeks
Team
2 engineers
Category
AI Chatbot
Where things stood
Larkspur's instructors were fielding the same conceptual questions repeatedly in office hours and forum threads, and students outside those hours had nowhere to get an answer beyond a static FAQ.
How we got there
Course-grounded RAG
Indexed every course's lecture transcripts, slides, and readings so the tutor answers from the actual course content, not generic knowledge.
Progress-aware responses
Tuned the assistant to reference where in the course a concept was covered, pointing students back to the right lecture instead of answering in isolation.
Escalation to instructors
Flagged questions the assistant couldn't answer confidently for instructor review instead of guessing, building a feedback loop for gaps in course content.
What changed
24/7
Tutor availability, up from office-hours only
~55%
Common questions answered without instructor involvement
30+
Courses using the AI tutor at launch
Tech stack
Claude API
Answers student questions grounded in each course's actual material.
RAG
Retrieves the specific lecture, slide, or reading a question relates to before answering.
Next.js
The course platform students and instructors use day to day.
PostgreSQL + pgvector
Stores course content and embeddings for retrieval without a separate vector store to sync.
TypeScript
Kept course content types consistent as the catalog grew to 30+ courses.
Vercel
Deploys platform updates without disrupting students mid-course.
