ITAS Multi-Agent Tutoring System Deployed in Quantum Computing Course
At Old Dominion University, researchers created and implemented the Intelligent Teaching Assistant System (ITAS), a multi-agent framework for LLM-based tutoring, within a graduate course on quantum computing over one semester. ITAS consists of three distinct layers: the teaching layer features a Spoke-and-Wheel configuration with three parallel specialist agents (Video, Code, Guidance) and a Synthesizer, along with an independent autograder that assesses the correctness and methodology of checkpoint submissions; the operational layer includes four Cloud Run microservices, utilizing Cloud SQL for session state and streaming interaction events to BigQuery via Pub/Sub; and the feedback layer contains a focused conversational agent that responds to instructor inquiries based on pseudonymized event streams for each lesson. This system tackles the Blind Instructor Problem, where LLM tutors gather more student data than instructors can utilize. The details are outlined in arXiv paper 2604.24808.
Key facts
- ITAS is a multi-agent tutoring system for LLM-based intelligent tutoring.
- It was used for a semester in a graduate quantum computing course at Old Dominion University.
- The teaching layer uses a Spoke-and-Wheel architecture with three specialist agents: Video, Code, Guidance.
- A Synthesizer follows the specialist agents in the teaching layer.
- A separate autograder evaluates correctness and approach of checkpoint submissions.
- The operational layer consists of four Cloud Run microservices.
- Session state is stored in Cloud SQL.
- Interaction events are streamed through Pub/Sub to BigQuery.
- The feedback layer is a narrow-scope conversational agent for instructor queries.
- It addresses the Blind Instructor Problem: LLM tutors accumulate more student data than instructors can access.
- The system is described in arXiv paper 2604.24808.
Entities
Institutions
- Old Dominion University
- arXiv
Locations
- Old Dominion University
- United States