ACE-TA: An Agentic Teaching Assistant for Grounded Q&A, Quiz Generation, and Code Tutoring

ACE-TA is an AI framework that combines large language models with three coordinated modules to provide automated educational support for programming students, including grounded question-answering, adaptive quiz generation, and interactive code tutoring with step-by-step guidance and sandboxed execution.

ACE-TA represents a practical application of agentic AI systems in educational technology, demonstrating how LLMs can be architecturally decomposed to handle specialized pedagogical tasks. The framework addresses a significant challenge in computer science education: scaling personalized instruction across diverse learning needs and conceptual depths. By separating concerns into retrieval-grounded Q&A, assessment generation, and interactive tutoring, the system reduces hallucination risks inherent in monolithic LLM deployments while maintaining coherence across the student experience.

This development builds on established trends in AI-assisted learning, where institutions increasingly seek cost-effective alternatives to human tutoring. However, ACE-TA's three-module architecture suggests maturation beyond simple chatbot implementations. The inclusion of sandboxed code execution differentiates it from text-only systems, enabling actual verification of student work—a critical capability for programming education where correctness matters.

For EdTech providers and institutions, ACE-TA's modular design offers a template for building domain-specific educational AI without reinventing core components. The quiz generation capability particularly enables institutions to maintain academic rigor while reducing instructor workload. The framework's reliance on pre-trained models keeps deployment costs manageable compared to custom-trained alternatives.

The critical metric going forward is educational efficacy: whether students using ACE-TA achieve better learning outcomes and retention compared to traditional or simpler AI-assisted approaches. Current limitations likely include handling of novel edge cases, context window constraints on large courses, and the pedagogical challenge of balancing guidance with autonomous problem-solving.

• →ACE-TA's three-module architecture (Q&A, quiz generation, code tutoring) represents a more sophisticated approach to AI-assisted education than single-purpose chatbots.
• →Sandboxed code execution with iterative feedback enables verification-based learning for programming courses.
• →The framework demonstrates how agentic routing reduces hallucination risks by specializing LLM tasks.
• →Educational efficacy and student outcome improvements remain the key metric for assessing real-world adoption.
• →The modular design creates a replicable template for other domain-specific educational AI applications.