Applied research
Most commentary on enterprise AI is opinion with a confident delivery. These are peer-reviewed instruments — published, indexed by Crossref, one of them already cited independently by researchers with no connection to me or my employer. Each one exists to be applied, not admired.
Published
Live DOIs, indexed and citable.
International Journal of Computer Applications, 187(143), 1–10
Enterprises govern AI before deployment — ethics reviews, data-sharing agreements, standards checklists — and then discover that the decisive failures all happen at runtime. A prediction has to be overridden. A model has to be rolled back. A decision has to be audited at 2am. Connectivity degrades and the system has to behave anyway.
The paper reframes governance as a runtime dependency and supplies the instrument: ten runtime controls, 232 coded source assignments across an 80-record evidence base, a six-family standards crosswalk, and four maturity levels with pass/fail questions and evidence requirements.
International Journal of Global Innovations and Solutions
Critical infrastructure does not get the luxury of a friendly operating environment. This reference architecture addresses AI deployed into grid systems under adversarial telemetry, constrained connectivity and continuous regulatory scrutiny — and specifies the security controls that must be hardened before AI is introduced at all.
Cited independently by Mir & Mir in the American Journal of Innovation in Science and Engineering at two separate points of their argument: for asset inventory and dependency analysis across cloud services, and for pre-AI control hardening.
2026 International Symposium on Networks, Computers and Communications (ISNCC), IEEE, pp. 1–7
Agents consume budget the way they consume tokens: continuously, and without asking permission. This framework moves cost governance inside the execution loop, so unit economics are enforced while agents run rather than reconciled after the invoice arrives.
Presented at ISNCC 2026 and published in IEEE Xplore. The reference implementation is open source, so the mechanism can be inspected and the results reproduced.
International Journal of Computer Applications, 187(138)
DocuMind: a deliberately lightweight document question-answering system built on TF-IDF, cosine similarity and rule-based extraction — a reminder that not every retrieval problem needs a frontier model or the bill that comes with one.
In publication
Peer review cleared and IEEE copyright transferred. Listed here with their real status rather than rounded up.
IICAIET 2026 · 8th IEEE International Conference on AI in Engineering and Technology, Kota Kinabalu
Token efficiency and cache correctness in hierarchical agent memory under controlled evaluation — the question of what an agent should remember, and what it costs to be wrong about that.
IEEE copyright transfer completed August 2026. Proceedings publication pending.
IEEE IEMCON 2026
Orchestration for disaster triage when connectivity degrades — deciding what must run at the edge, what can wait for the cloud, and how the system stays correct across the partition.
Camera-ready submitted. Proceedings publication to follow.
Direction of travel
Three manuscripts are currently under review. They are listed for transparency and are not presented as published work — but they show where the architecture is heading.
A reference architecture and reproducible stress test · ICogSys 2026
Computer vision and NLP in a production advisory chatbot · IEEE-CICON 2026
Bi-LSTM with IRSA-based feature selection · Engineering (Elsevier)
Full bibliographic records are maintained on ORCID and Google Scholar. Publication status was verified against Crossref and publisher pages in September 2026.
Next step
These frameworks were designed to be used. An architecture review maps your environment against them and returns findings a board can act on.
