Personalized Learning Diagnosis Approach with Multi-Agent Collaboration and Four-Layer Anti-Hallucination Mechanism

DOI: https://doi.org/10.62517/jbdc.202601320

Author(s)

Zhengxuan Luo¹,*, Huaiming Fan¹, Wen Chen¹, Mingming Gong²

Affiliation(s)

1Henan University of Technology, Zhengzhou, Henan, China 2iFLYTEK Co., Ltd., Hefei, Anhui, China *Corresponding Author

Abstract

Aiming at uneven learning foundations, opaque teaching workflows and hallucination-reasoning defects of single-LLM solutions in university AI education, this paper presents ZhiCheng, a personalized learning-diagnosis system with the “1+7+8” multi-agent architecture. Built upon the dual-brain functional partition, seven core agents are scheduled by an Orchestrator, while eight extended nodes constitute a quality-control loop. A four-tier anti-hallucination pipeline cuts factual error rate from 15%-25% to below 1%. Under 1250 concurrent users, the system reaches 842 TPS, and index tuning reduces core-query P95 latency by 60%. Experiments, ablation studies and closed-loop tests validate the architecture superiority and system effectiveness.

Keywords

Multi-Agent Collaboration; Langgraph Orchestration; Dual-Brain Division-Of-Labor Model; Illusion Prevention; Learning Profile; Personalized Learning; Retrieval-Enhanced Generation; Stress Test

References

[1] Unesco. Guidance for generative AI in education and research. Paris: UNESCO, 2023. [2] Kasneci E, Sessler K, Kuchemann S, et al. ChatGPT for good? On opportunities and challenges of large language models for education. Learning and Individual Differences, 2023, 103: 102274. [3] Ji Z, Lee N, Frieske R, et al. Survey of hallucination in natural language generation. ACM Computing Surveys, 2023, 55(12): 1-38. [4] Huang L, Yu W, Ma W, et al. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. arXiv preprint arXiv:2311.05232, 2023. [5] Fastapi Contributors. FastAPI framework documentation. [2026-08-06]. https://fastapi.tiangolo.com/. [6] LangGraph Contributors. LangGraph: Building stateful, multi-actor applications with LLMs. (2024-01-15). https://github.com/langchain-ai/langgraph. [7] Sqlite Consortium. Write-ahead logging. [2026-08-06]. https://www.sqlite.org/wal.html. [8] Hong S, Zhuge M, Chen J, et al. MetaGPT: Meta programming for a multi-agent collaborative framework//The Twelfth International Conference on Learning Representations. 2024. [9] React Flow Contributors. React Flow: A customizable React component for building node-based editors and interactive diagrams. [2026-08-06]. https://reactflow.dev/. [10] Chroma Contributors. Chroma documentation. [2026-08-06]. https://docs.trychroma.com/. [11] Asai A, Wu Z, Wang Y, Et al. Self-RAG: Learning to retrieve, generate, and critique through self-reflection//The Twelfth International Conference on Learning Representations. 2024. [12] Deepseek-Ai. Deepseek-V3 technical report. arXiv:2412.19437, 2024[2026-08-06]. https://arxiv.org/abs/2412.19437. [13] Zheng L, Chiang W L, Sheng Y, et al. Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena//Advances in Neural Information Processing Systems. 2023, 36: 46595-46623. [14] Li J, Cheng X, Zhao X, et al. HaluEval: A large-scale hallucination evaluation benchmark for large language models//Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. Singapore: Association for Computational Linguistics, 2023: 6449-6464. [15] Apache Software Foundation. Apache JMeter user manual. [2026-08-06]. https://jmeter.apache.org/usermanual/.