Article Details

Abstract

This paper proposes a localized intelligent teaching system based on LangChain and DeepSeek-7B-R1, aiming to solve the challenges of asynchronous programming comprehension and DOM operation error detection in JavaScript and jQuery teaching. To address the conflicts among semantic understanding, privacy security, and hardware requirements in traditional solutions, this paper designs a dual-engine architecture (rule pre-screening + LLM fine-tuning) that integrates dynamic AST parsing and an enhanced RAG retrieval mechanism, constructing a teaching Knowledge Graph covering 57 types of typical errors. Experiments demonstrate that the system achieves an average response time of 2.8 seconds on NVIDIA RTX 3060 device, and the teaching suggestion accuracy reaching 91.2%, with a 43.7% improvement over conventional LSTM approaches. Deployment in corporate internal training scenarios verifies its capability to enhance learner problem-solving efficiency by 41% while reducing teacher manual intervention by 76%. This research provides the first high-precision and localized intelligent teaching framework supporting consumer-grade hardware for the programming education field, with open-source core components to lower development barriers.

Keywords

localized intelligent teaching system; LangChain rule engine; DeepSeek-7B-R1 fine-tuning; RAG mechanism; JavaScript Teaching Knowledge Graph; AST dynamic parsing; consumer-grade hardware deployment; programming education efficiency optimization

References

1. [1] WANG Z Q，LIU J，ZHANG S K，et al. Poisoned LangChain: Jailbreak LLMs by LangChain [J/OL].arXiv:2406.18122 [cs.CL].[2025-01-20].https://arxiv.org/html/2406.18122v1.
2. [2] HU E J，SHEN Y L，WALLIS P，et al. LoRA: Low-rank Adaptation of Large Language Models [J/OL].arXiv:2106.09685 [cs.CL].[2025-01-21].https://arxiv.org/abs/2106.09685?context=cs.LG.
3. [3] LangChain Team. Vodafone Transforms Data Operations with AI Using LangChain and LangGraph [EB/OL].[2025-01-16].https://blog.langchain.dev/customers-vodafone/.
4. [4] LangChain Team. LangChain Documentation [EB/OL].[2024-05-01].https://python.langchain.com/docs/.
5. [5] ZHOU P，PUJARA J，REN X，et al. Self-discover: Large Language Models Self-compose Reasoning Structures [J/OL].arXiv:2402.03620 [cs.AI].[2025-01-16].https://arxiv.org/abs/2402.03620.
6. [6] SARTHI P，ABDULLAH S，TULI A，et al. RAPTOR: Recursive Abstractive Processing for Tree-organized Retrieval [J/OL].arXiv:2401.18059 [cs.CL].[2025-01-03].https://arxiv.org/abs/2401.18059.
7. [7] LEWIS P，PEREZ E，PIKTUS A，et al. Retrieval-augmented Generation for Knowledge-intensive NLP Tasks [J/OL].arXiv:2005.11401 [cs.CL].[2025-02-06].https://arxiv.org/abs/2005.11401.
8. [8] VIDIVELLI S，RAMACHANDRAN M，DHARUNBALAJI A. Efficiency-driven Custom Chatbot Development: Unleashing LangChain, RAG, and Performance-optimized LLM Fusion [J].Computers，Materials & Continua，2024，80（8）：2423-2442.
9. [9] Min Rui, Jiang Dan, Wang Yonggang Research on Medical Question Answering System Based on LangChain and Large Model [J]. Computer Science and Applications, 2025, 15 (2): 33-43
10. [10] LANG G，TRIANTORO T，SHARP J H. Large Language Models as AI-powered Educational Assistants: Comparing GPT-4 and Gemini for Writing Teaching Cases [J].Journal of Information Systems Education，2024，35（3）：390.

About the authors

Liu Anqian (1980-), male, Han ethnicity, from Dalian, Liaoning Province, China, teacher, master's degree, research interests: artificial intelligence, low code platforms, big data analysis, vocational education, robotics.