Bobcat AI Tutor

RAG-based AI tutoring system integrated with Texas State course management APIs, serving grounded 24/7 student support and reducing TA workload by 50%.

A retrieval-augmented AI tutoring system built during my graduate research assistantship at Texas State University. Integrates with the course management system APIs to ground responses in actual course content — syllabi, lecture notes, assignments — reducing hallucination and instructor workload by 50%.

The problem with generic LLM tutors

Off-the-shelf LLM assistants don’t know anything about the specific course. Students get generic answers that may contradict what their instructor actually expects. The fix is retrieval: every response is grounded in the actual course materials for that student’s specific section.

Pipeline

1. Course materials ingested from Canvas API (PDFs, pages, assignments)
2. Chunked by section with metadata tags (course, week, content type)
3. Embedded with text-embedding-3-small and stored in ChromaDB per course
4. At query time: retrieve top-k chunks for the student’s course, pass to Claude for synthesis
5. Responses cite the source chunk so students can verify

Outcome

Deployed to graduate-level CS and data science courses. Reduced TA office hour load by 50% in pilot semester.