ITAS: A Multi-Agent Architecture for LLM-Based Intelligent Tutoring

Old Dominion University researchers have developed ITAS (Intelligent Teaching Assistant System), a multi-agent architecture for LLM-based tutoring that ran for a full semester in a graduate quantum computing course. The system addresses the gap between easy-to-build notebook prototypes and production-ready tutoring tools. ITAS comprises three layers: the teaching layer uses a Spoke-and-Wheel design with three parallel specialist agents (Video, Code, Guidance) feeding into a Synthesizer, plus a separate autograder that evaluates both correctness and approach. The operational layer runs on four Cloud Run microservices with session state in Cloud SQL and interaction events streamed through Pub/Sub to BigQuery.

The feedback layer tackles what the authors call the Blind Instructor Problem—LLM tutors accumulate more student data than instructors can access through routine channels. A narrow-scope conversational agent answers instructor questions over per-lesson pseudonymized event streams. In a five-student pilot, the teaching layer handled 334 chat turns without task-boundary hallucinations, the operational layer captured 10,628 events across five modules, and the feedback layer surfaced two findings the instructor acted on mid-semester. The researchers don't claim generalizability but position ITAS as a workable end-to-end answer for running LLM-based intelligent tutoring in real courses.