DeepTutor: An Agentic Framework for Personalized Tutoring with LLMs
DeepTutor is an open-source framework designed for personalized tutoring that overcomes the shortcomings of traditional tutoring systems and RAG-augmented methods. It features a hybrid personalization engine that merges static knowledge grounding with a dynamic multi-resolution memory, allowing it to transform interaction history into developing learner profiles. The system operates within a closed tutoring loop, linking citation-based problem-solving with questions generated based on difficulty levels. Furthermore, the personalization substrate facilitates collaborative writing, deep research involving multiple agents, and integration options. This framework seeks to enhance the provision of tailored, guided feedback in the educational applications of large language models.
Key facts
- DeepTutor is an agent-native open-source framework for personalized tutoring.
- It uses a hybrid personalization engine with static knowledge grounding and dynamic multi-resolution memory.
- The system distills interaction history into a continuously evolving learner profile.
- It features a closed tutoring loop coupling citation-grounded problem solving with difficulty-calibrated question generation.
- The personalization substrate supports collaborative writing and multi-agent deep research.
- The framework addresses limitations of conventional tutoring systems that lack adaptation to individual learners.
- Existing RAG-augmented systems fall short in delivering personalized, guided feedback.
- The paper is available on arXiv with ID 2604.26962.
Entities
Institutions
- arXiv