Reproduce “DeepTutor: Towards Agentic Personalized Tutoring” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Bingxi Zhao, Jiahao Zhang, Xubin Ren, Zirui Guo, Tianzhe Chu, Yi Ma, Chao Huang
DeepTutor is an open-source agentic framework that integrates personalized problem tutoring with difficulty-calibrated question generation, driven by a dynamic learner model. It demonstrated an average 10.76% improvement in overall tutoring quality on the new student-centric TutorBench benchmark compared to a naive baseline.
Key insights
- Integrating a dynamic learner memory (DPM) and static knowledge grounding (SKG) within a closed-loop, agentic framework is crucial for achieving truly personalized and adaptive LLM-based tutoring.
- The "investigate–solve–write" scaffold for problem solving demonstrates general utility, improving performance across various LLM backbones on diverse agentic reasoning benchmarks, even without personalization.
- Evaluating personalized tutoring systems requires student-centric benchmarks like TutorBench that can interactively assess multi-turn adaptive behavior grounded in specific course materials and learner profiles.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.