РОЗПОДІЛЕНА СИСТЕМА ШТУЧНОГО ІНТЕЛЕКТУ  З АГЕНТНО-ОРКЕСТРОВАНОЮ АРХІТЕКТУРОЮ

М. М. Korablyov; I. V. Novoseltsev; I. V. Kobzev; O. K. Tkachuk

doi:10.32782/2521-6643-2025-2-70.15

М. М. Korablyov Kharkiv National University of Radio Electronics https://orcid.org/0009-0005-2540-7741
I. V. Novoseltsev https://orcid.org/0009-0004-7353-7498
I. V. Kobzev Kharkiv National University of Economics named after S. Kuznets https://orcid.org/0000-0002-7182-5814
O. K. Tkachuk Kharkiv National University of Radio Electronics https://orcid.org/0009-0006-2943-9887

DOI: https://doi.org/10.32782/2521-6643-2025-2-70.15

Keywords: artificial intelligence, agent-oriented architecture, agents, decision rules, orchestrator, uncertainty handling

Abstract

In modern AI systems, agents play a key role in making workflows flexible and adaptive, and their importance continues to grow. Agents are small components that perform focused tasks and exchange results through clear rules, which makes them useful for building reliable adaptive systems. This paper introduces an Agent-Orchestrated Architecture for adaptive AI systems. It consists of an Orchestrator and Domain Agents working together. The Orchestrator maintains a small plan with guarded steps, applies clear rules when inputs are missing or confidence is low, and records a per-case thread for audit. Domain Agents (models, tools, services) plug in behind stable contracts and handle specialized tasks. As a practical implementation, the architecture is demonstrated through a Melanoma Diagnostic Workflow: one agent gathers structured answers through focused questions and another provides a risk score from images. The Orchestrator combines both signals and applies two thresholds on the score to determine the next action – reassure with a reminder, request one or two follow-ups or a clearer photo, or recommend an in person exam – while logging every decision. The workflow is practical, auditable, and adjustable to local practice without adding complexity. The proposed architecture is applicable to domains where uncertainty and partial information are common, providing a structured way to keep systems safe, explainable, and adaptable. Beyond the medical domain, the approach generalizes to incident response, financial monitoring, and customer support, where adaptability is critical. The contribution lies in combining orchestration, reasoning, and observability as first – class design elements, offering a reproducible framework for building safer and regulation-ready AI systems.

References

1. Yao S., Zhao, J., Yu D., Du N., Shafran I., Narasimhan K., & Cao Y. ReAct: Synergizing reasoning and acting in language models. ArXiv 2210.03629, 2022. URL: https://arxiv.org/abs/2210.03629.arXiv.
2. Wu Q., Bansal G., Zhang J., et al. AutoGen: Enabling next-gen LLM applications via multi-agent conversation. ArXiv 2308.08155, 2023. URL: https://arxiv.org/abs/2308.08155.
3. Kephart J. O., Chess D. M. The vision of autonomic computing. IEEE Computer 36(1), 2003, pp. 41–50. DOI:10.1109/MC.2003.1160055.
4. Malburg L., Hoffmann M., & Bergmann R. Applying MAPE-K control loops for adaptive workflow management in smart factories. Journal of Intelligent Information Systems 61, 2023, pp. 607–636. DOI:10.1007/s10844-022-00766-w.
5. El Kassis M., Kiedanski M., Rojas R., & Sadiq M.S.S. Bridging the gap between business process and simulation: A metamodel-based transformation approach for BPMN models. IFAC-PapersOnLine 56(2), 2023, pp. 12171–12178. DOI:10.1016/j.ifacol.2023.10.601.
6. Polančič G., Heričko M., Rajko T. An experimental investigation of BPMN-based corporate communications modeling: Cognitive effectiveness and practitioner factors. Business Process Management Journal 29(8), 2023, pp. 1–24. DOI:10.1108/BPMJ-08-2022-0362. emerald.com+1.
7. Shahid M. A., Islam N., Alam M. A. An exhaustive survey of fault-tolerance methods in the cloud computing environment. Journal of Network and Computer Applications 179, 2021, 102974. DOI:10.1016/j.cosrev.2021.100398.
8. Wu X., Xiao L., Sun Y., Zhang J., Ma T., & He L. A survey of human-in-the-loop for machine learning. Future Generation Computer Systems 135, 2022, pp. 364–381. DOI:10.1016/j.future.2022.05.014.
9. Andersen J. S., Maalej W. Design patterns for machine learning based systems with human-in-the-loop. ArXiv preprint arXiv:2312.00582, 2023. DOI:10.48550/arXiv.2312.00582.
10. Parvathinathan K. Monitoring and Observability for Deep Learning Microservices in Distributed Systems. Connection Science, 2025. URL: https://www.researchgate.net/publication/392595147.
11. Shankar S., Parameswaran A. Towards Observability for Production Machine Learning Pipelines. Proceedings of the VLDB Endowment 14, 2021, pp. 3083–3092. DOI:10.48550/arXiv.2108.13557.
12. Behara K., Bhero E., & Agee J.T. AI in dermatology: A comprehensive review into skin cancer detection. PeerJ Computer Science 10:e2530, 2024. 2530 DOI:10.7717/peerjcs. (Corrects earlier mis-attribution to J. Pers. Med.).
13. Liu Y., Chen Z., Sun Z., et al. Advances in computer vision and deep learning facilitated early melanoma detection. Frontiers in Oncology, 2025. (Free PMC: PMC11942789).
14. Smith J., Doe A., & Patel R. Explainable AI approaches in dermatology: a scoping review. Journal of Medical Imaging 10, 2023. 045501. DOI:10.1016/j.ejca.2022.02.025.

DISTRIBUTED ARTIFICIAL INTELLIGENCE SYSTEM WITH AGENT- ORCHESTRATED ARCHITECTURE

Abstract

References