PIOAGENT: A HYBRID FINITE-STATE FRAMEWORK FOR DETERMINISTIC ORCHESTRATION OF LLM AGENTIC WORKFLOWS
Elmi Əsərlər · 25.06.2026 · 35 (1) · 47-53
DOI: 10.61413/qcir6462 ↗Kurz gefasst
Fachartikel in der Zeitschrift Elmi Əsərlər, online erschienen am 25.06.2026. Verfasst von Alakbar Taghi Valizada.
Abstract (Originalsprache)
Large-language-model (LLM) agents built on free reasoning loops produce non-reproducible control flow, may run without bound, and fail in brittle ways, which makes them hard to deploy in regulated, latency-bound conversational systems such as contact centres and government service desks. This paper presents PioAgent, an orchestration framework that separates a deterministic finite-state control layer from the stochastic computation performed inside nodes. Each phase runs either as a Cyclic Directed Graph for controlled, stateful loops with human-in-the-loop steps and retries, or as an Event-Driven Graph for retrieval-augmented and data-availability-triggered pipelines. The framework adds durable snapshots, declarative retry and fallback policies, an exactly-once effect outbox, structured-output streaming, and a model-agnostic adapter layer, all expressed in a declarative configuration. A prototype has been implemented, and an evaluation against ReAct-style loops, AutoGen, and LangGraph on a contact-centre workload is presented as a set of falsifiable predictions tied to the design: perfectly reproducible control trajectories under fixed decoding, routing accuracy that resists degradation as the tool catalogue grows, single-digit-to-low-double-digit orchestration overhead, and logarithmic-cost fault localization. The work shows that confining stochasticity to node interiors yields agentic systems that are simultaneously expressive, auditable, and recoverable.
Zitieren
APA
Valizada, A. T. (2026). PIOAGENT: A HYBRID FINITE-STATE FRAMEWORK FOR DETERMINISTIC ORCHESTRATION OF LLM AGENTIC WORKFLOWS. Elmi Əsərlər, 35(1), 47-53. https://doi.org/10.61413/qcir6462
Vancouver
Valizada AT. PIOAGENT: A HYBRID FINITE-STATE FRAMEWORK FOR DETERMINISTIC ORCHESTRATION OF LLM AGENTIC WORKFLOWS. Elmi Əsərlər. 2026;35(1):47-53. doi:10.61413/qcir6462
BibTeX
@article{Valizada2026,
author = {Valizada, Alakbar Taghi},
title = {PIOAGENT: A HYBRID FINITE-STATE FRAMEWORK FOR DETERMINISTIC ORCHESTRATION OF LLM AGENTIC WORKFLOWS},
journal = {Elmi Əsərlər},
year = {2026},
volume = {35},
number = {1},
pages = {47--53},
doi = {10.61413/qcir6462}
}
EUVIMED