Governance-as-Code for Deterministic, Reliable and Reproducible AIoT Systems in Smart Buildings

Young Im Cho; Saydirasulov Norkobi; Davronbekov Dilmurod; Dusmurod Kilichev

doi:10.65164/46gydd66

Mualliflar

Young Im Cho Department of Computer Engineering, Gachon University, Seongnam-si, Republic of Korea ##default.groups.name.author##
Saydirasulov Norkobi Department of Computer Engineering, Gachon University, Seongnam-si, Republic of Korea; Department of Technologies of Mobile Communication, Tashkent University of Information Technologies named after Muhammad Al-Khwarizmi, Tashkent 100200, Uzbekistan; ##default.groups.name.author##
Davronbekov Dilmurod Department of Technologies of Mobile Communication, Tashkent University of Information Technologies named after Muhammad Al-Khwarizmi, Tashk ##default.groups.name.author##
Dusmurod Kilichev Department of Information Security, National University of Uzbekistan, Tashkent 100174, Uzbekistan ##default.groups.name.author##

DOI:

https://doi.org/10.65164/46gydd66

Kalit so‘zlar:

AIoT; governance-as-code; smart buildings; deterministic orchestration; tail latency; MTTR; reproducibility; cyber-physical systems; Industry 5.0

Abstrak

Reliable and reproducible operation remains difficult in Artificial Intelligence of Things
(AIoT) systems for smart buildings, where rapid software updates must coexist with safety, privacy,
resilience and energy-efficiency requirements. This article presents a shortened journal version of
a governance-as-code framework that embeds executable policies into both CI/CD pipelines and
runtime control. The proposed monolithic-modular architecture centralizes admission control,
intent routing, capability resolution, scheduling, health management and rollback in a deterministic
orchestrator, while allowing independently versioned AI service modules through standardized
contracts. A normalized device layer abstracts heterogeneous building protocols such as BACnet/IP,
Modbus/TCP, OPC-UA and MQTT, preventing direct module-device interaction and supporting
coordinated recovery. The deployment problem is formulated as a constrained multi-objective
trade-off over tail latency, mean time to repair (MTTR), availability, energy-normalized comfort
and governance overhead. Evaluation follows a stage-consistent pipeline spanning simulation,
hardware-in-the-loop experiments and a production-like prototype. Across these stages, the
proposed architecture reduces P90 action latency by approximately 40%, reduces MTTR by about
60%, approaches 99.9% availability under fault injection, improves energy-normalized comfort by
approximately 15% and keeps governance overhead within 3-5%. These findings suggest that
governance-as-code can transform compliance from static documentation into a continuously
enforced operational mechanism for trustworthy and sustainable AIoT systems.