DeepDomain AI: System Design and Validation for Enterprise AI Talent Training

DOI: https://doi.org/10.62517/jbdc.202601321

Author(s)

Pengyue He, Suya Cheng*, Weili Wang, Lulu Ma, Yutong Li

Affiliation(s)

School of Artificial Intelligence and Big Data, Henan University of Technology, Zhengzhou, Henan, China *Corresponding Author

Abstract

Enterprise AI roles evolve faster than conventional course catalogs, while direct use of large language models (LLMs) for training-content generation introduces risks such as unsupported claims, non-executable code, and incomplete audit trails. This paper presents DeepDomain AI, a prototype platform for personalized enterprise AI training. Built with React 18 and FastAPI, the platform compares learner evidence with a structured catalog of 15 AI roles, identifies readiness gaps, and organizes role-oriented training paths. Its resource workflow produces foundation-oriented and project-oriented candidates, records structured review results, and applies a publish/correct/reject quality gate. Pydantic schemas constrain model outputs, and a Docker sandbox scores code through task-specific executable assertions rather than source-code keywords. Functional tests cover job analysis, path generation, knowledge recommendation, state persistence, and sandbox execution. In an offline evaluation of 45 constructed learner profiles, dynamic ranking achieved a Top-3 hit rate of 93.33%; on the 30 same-direction profiles used for comparison, the static-label baseline achieved 80.00%. A 20-sample threshold-consistency test produced a 100% interception rate for rule-violating resources and a 0% false-positive rate for rule-compliant resources. In this study, the system was tested with constructed learner profiles rather than data collected from employees. The results therefore describe how the implemented rules and ranking method performed in our test setting; they do not yet show whether the generated resources improve learning or meet expert expectations in practice.

Keywords

AI Workforce Training; Multi-Agent Collaboration; Learner Profile; Job Knowledge Graph; Resource Quality Gate; Hallucination Control

References

[1] Liu Y, Li W, Wang Y, et al. ISFJ-RAG: Interventional Suppression of Hallucinations via Counter-Factual Joint Decoding Retrieval-Augment Generation. Big Data and Cognitive Computing, 2026, 10(2): 56. [2] Sun X, Hu Y, Gao X, et al. A cooperative multi-agent optimization approach for task offloading in vehicular edge computing systems. Discover Computing, 2025, 28(1): 348. [3] Li H, Xing W, Li C, et al. I hear you: Enhancing middle school mathematics engagement and learning outcomes through teachable agent voice design. Computers & Education, 2026, 254: 105725. [4] Saini G, Mishra A, Jadon S S.A neural-guided artificial bee colony with dynamic clustering and adaptive quality gating for global optimization. Expert Systems with Applications, 2026, 317: 131874. [5] Ciroku F, de Berardinis J, Kim J, et al. RevOnt: Reverse engineering of competency questions from knowledge graphs via language models. Journal of Web Semantics, 2024, 82: 100822. [6] Yang B, Shen Z. Knowledge graph construction and talent competency prediction for human resource management. Alexandria Engineering Journal, 2025, 121: 223-235. [7] Ouared A, May M, Piau-Toffolon C, et al. TracePath: Modeling and Analyzing Competency Trajectories With Graph‐Based Learning Analytics Over a Hybrid Polystore. Concurrency and Computation: Practice and Experience, 2026, 38(1): e70508. [8] Li G, Yuan C, Kamarthi S, et al. Data science skills and domain knowledge requirements in the manufacturing industry: A gap analysis. Journal of Manufacturing Systems, 2021, 60: 692-706. [9] Deshmukh P, Samrat H, Pervin N. Personalized learning path generation using large language models and sequential recommendations. Data & Knowledge Engineering, 2026, 164: 102595. [10] Liu J, Yan C, Liu W, et al. Research on learner “emotion-behavior-ability” characteristics based on MOOC online education user profiles. Information Processing and Management, 2025, 62(3): 104026. [11] Le H V, Ngo Q D. V-Sandbox for Dynamic Analysis IoT Botnet. IEEE ACCESS, 2020, 8: 145768-145786. [12] Ivanchenko Y, Karpinski M, Ryzhakov M, et al.An Agent-Based Model of a Controlled Detonation System for Sandbox Analysis of Suspicious Software. Electronics, 2026, 15(11): 2348.