As AI workloads squeeze electrical and thermal margins, one engineer’s research proposes a way for specialized software roles to coordinate across domains while an independent safety check and a qualified person keep the final say.
AI accelerators have changed what a data center asks of its electrical and cooling plant. Racks that draw far more power per footprint compress the margins that UPS systems, medium-voltage switchgear, generators and cooling equipment once had to spare. Facilities designed around N+1 or 2N redundancy can now run closer to their thermal, capacity and protection limits, which raises the cost of any single misstep.
1
Raja Kumara Swamy Donthula, a data-center electrical and commissioning engineer based in the Atlanta area, argues that the answer is not a better dashboard for each system in isolation. In a research manuscript on safety-critical data-center electrical operations, he proposes a decision support framework in which specialized software roles reason across domains together, under explicit uncertainty, with an independent safety check and a qualified person in the decision path.
One facility, several conversations
In practice, power, cooling and commissioning are often managed by different teams, tools and checklists. Yet the physical plant does not respect those boundaries. A load increase on the IT side shows up as less electrical headroom and more heat to remove. Taking a UPS string out for maintenance changes the redundancy that the rest of the plant depends on. A gap in commissioning records changes how much of the as-built system anyone can vouch for.
The research starts from that coupling. The practical question is how engineering teams can coordinate these domains when the data is imperfect, the margins are thin and the consequences of a wrong action are high.
Why alarms and thresholds are not enough
Most building-management and infrastructure-management workflows remain built around alarms and thresholds. They are good at reporting that a value has crossed a line. They are less suited to weighing power quality, breaker and relay status, battery health and thermal state at the same time.
Newer agentic AI assistants could fill part of that gap, but they bring their own risk: a fluent recommendation that carries no calibrated confidence and no independent check against interlocks and protective coordination. That is the failure mode the framework is designed to contain.
Specialized roles, one shared record
The proposal divides the work among six roles. One tracks equipment condition and aging. One maintains a live model of the electrical one-line diagram and the redundancy state. One fuses the evidence into a calibrated risk score. One ranks corrective actions, putting safety ahead of efficiency. One independently verifies every candidate action. One presents verified, explained options to the operator.
All of them write to a shared, append-only record, so that a reviewer can later see what each role knew, how sure it was and what it recommended. The framework is deliberately agnostic about how each role is implemented; it specifies the interfaces and the safety logic rather than a particular learning algorithm.
2
Conceptual diagram of the proposed decision-support framework; a research concept, not a deployed system.
Uncertainty travels with every message
Each message between roles carries a state estimate, an uncertainty field, a data-quality score and its provenance, including source identifiers, timestamps and model version. The data-quality score takes the weakest of four components (freshness, completeness, channel health and consistency with the one-line model) rather than an average, so a stale input cannot hide behind healthy ones.
Where a role cannot produce a probabilistic uncertainty estimate, the field is explicitly marked null. The framework states that a null must never be read as high confidence.
An independent gate that fails closed
The safety role is separate from the roles that propose actions. It checks each candidate against interlock, arc-flash and protective-coordination constraints that qualified personnel derive from approved one-line diagrams, protection studies and switching procedures.
If topology, telemetry freshness, protection-study validity or interlock state cannot be confirmed, it withholds clearance and escalates. A favorable risk trade-off never substitutes for its approval. Even fallback options must pass the same check before reaching an operator.
Human authority scaled to consequence
The framework ties how much a system may do on its own to the consequence and reversibility of the action. High consequence, high uncertainty or low reversibility keeps a recommendation advisory only. Moderate-consequence actions that can be reversed through a pre-approved recovery procedure can move to approve-to-execute. Only low-consequence, bounded and pre-
3
approved actions inside a validated envelope are candidates for limited automatic execution, with review afterward.
Consequence and reversibility act as hard ceilings. Low uncertainty alone cannot raise the level of autonomy.
What the framework deliberately does not do
The manuscript draws firm boundaries around automation. The framework is decision support, not autonomous control: most roles are advisory and non-executing, the safety role only grants or withholds clearance, and the operator-support role presents and routes recommendations. Bounded automatic execution explicitly excludes protective-relay actions, energization, breaker or automatic-transfer-switch transfers, generator paralleling, bypass operations and changes to lockout-tagout or arc-flash exposure.
If authorization is unavailable, the action stays blocked and escalates; there is no timeout-to execute path. The framework also does not calculate arc-flash incident energy itself. It consumes facility-approved study results, performed under IEEE Std 1584 and combined with NFPA 70E work-practice constraints, as external safety inputs.
A cross-domain test case: compute, cooling and a shared bus The coordination idea is easiest to see in one of the manuscript’s simulated scenarios. A compute load and a cooling load share an electrical bus, and an injected event derates that bus for part of the run. In the uncoordinated baseline, each load submits its own locally sensible request with no view of the other, so the shortfall is spread across both and nothing protects a minimum level of cooling.
In the coordinated configuration, the response-ranking role curtails compute first, and the safety role enforces a floor on the cooling that is served, cutting cooling only as a last resort. In that simulation, both bus-capacity violations and cooling-floor breaches fell to zero.
The author is careful about what that does and does not prove. Because the coordinated arm was given a cooling floor that the baseline lacked, a zero-breach result is guaranteed by construction. When the manuscript added a deterministic real-time rule that used the same bus limit, floor and telemetry, that simple rule matched the coordinated arm on every matched trial. The manuscript flags this as an important negative result for any claim that agentic reasoning itself adds value in that scenario. In other words, there the explicit constraint, not the sophistication of the reasoning, does the work.
How the framework would be tested
The manuscript formalizes risk, uncertainty and the safety gate mathematically, and it specifies, but does not yet carry out, a fuller evaluation on a physics-informed digital twin. The twin would combine an electrical one-line and power-flow model with thermal and cooling co-simulation and
4
a fault-injection engine. Four scenario families are planned: UPS degradation, medium-voltage feeder overload with the loss of a feeder, cooling-driven derating after a loss of cooling capacity, and integrated systems test (IST) sequencing with stale breaker status.
Comparisons would include a threshold-and-rule baseline, a single-agent optimizer and a strong non-agentic supervisory controller, along with ablations that remove uncertainty propagation, the independent safety gate or the human feedback loop. Toolchain, solver settings, topology, ratings and random seeds are frozen before any comparative run, and the statistical plan, which calls for at least 30 matched replications per condition, is set in advance. Metrics include detection latency, mitigation time, calibration error, unsafe-action rate, veto rate and residual constraint violations.
What the first-pass simulations do and do not show
Ahead of that study, the manuscript reports five small pilot simulations that use illustrative parameters rather than facility measurements. The author describes the results as preliminary and non-conclusive, and several are mixed. A predictive feeder controller, for example, did not improve redundancy-margin compliance at its default setting; the manuscript treats that as a design lesson about how the target ceiling trades compliance against load shed. In the cooling scenario, predictive throttling reduced temperature violations only partly, because some degraded cases cannot be recovered by throttling alone.
A commissioning-sequencing pilot found that most of its gain came from simply reassigning crews when tasks slipped, not from the more sophisticated prioritization layered on top of it.
That candor is part of the argument. The full digital-twin study remains future work, and no deployment on live data-center equipment is claimed.
Key takeaways for operators
- Treat power, cooling and commissioning as one coupled system when evaluating any AI assisted recommendation.
- Require each recommendation to state its evidence, its uncertainty and the quality of the data behind it, and treat missing uncertainty as unknown, not safe.
- Keep an independent check that can say no, and that fails closed when data or topology cannot be confirmed.
- Match the level of automation to consequence and reversibility, not to model confidence alone.
- Benchmark AI-assisted tools against a strong non-AI baseline; in some scenarios a simple, explicit rule may deliver most of the benefit.
- Keep commissioning and as-built records current; decision support is only as reliable as the evidence it reads.
5
For data centers adding capacity and intelligence at the same time, the more useful question may not be whether AI can recommend an action. It may be whether the surrounding engineering system can show why, check the recommendation independently and hand the decision to the right person.
About the research
Raja Kumara Swamy Donthula, P.Eng., Senior Member, IEEE is a data-center electrical and commissioning engineer based in the Atlanta area. The framework discussed here is described in his manuscript, “Uncertainty- and Risk-Aware Multi-Agent Decision-Support Framework for Safety-Critical Data-Center Electrical Operations.
6
TECHANNOUNCER FEATURE — FINAL EDITORIAL & TECHNICAL REVIEW
Appendix to the original six-page article | Review date: 24 October 2026 | Prepared for Team A | INTERNAL REVIEW COPY — not for publication
Article: “Cross-Domain Coordination Under Engineering Constraints: Why AI Data Centers Need Human-Supervised Decision Support Across Power, Cooling and Commissioning”
Recommendation: CONDITIONAL APPROVAL AFTER TARGETED CORRECTIONS. The feature is technically thoughtful, accessible and largely consistent with the underlying research manuscript. It is suitable to move toward an editorial submission after the factual designation, publication metadata and specific wording/layout edits below are addressed. No fundamental rewrite or additional experiment is required for this media feature.
Basis and limits of this review
Reviewed the supplied six-page article, including its photograph and conceptual diagram, against the eight-page working research manuscript, “Uncertainty- and Risk-Aware Multi-Agent Decision-Support Framework for Safety-Critical Data-Center Electrical Operations.” This is a technical editorial fact-check of a media feature; it is not independent academic peer review of the research, a replication of the simulations or confirmation of publication by TechAnnouncer.
Prioritized editorial checklist
Priority / page Finding Team A action
BLOCKER | pp. 1, 6 “P.Eng.” appears after the researcher’s name although the BC licensing assessment
was last reported as still pending.
Remove P.Eng. from both locations until formal registration has been granted; confirm title against the regulator before publishing.
REQUIRED | article metadata
The supplied PDF has no identifiable publication byline, publication date or final article URL.
For a public release, add/retain publisher name, journalist/byline, dated publication, category and canonical online link. Preserve the published version.
REQUIRED | p. 6 The full manuscript title is opened with a quotation mark but not closed. The page also
has large unused space.
Close quotation and sentence. Consider fitting the “About the research” note on page 5 if the outlet supports it.
REQUIRED | pp. 2–3
IMPORTANT | pp. 2–5
Six separate agent roles are described in prose, but only five boxes appear because Prognostics and Topology/Capacity are combined.
Several descriptions read like an implemented operational system, while the paper describes proposed interfaces and pilot simulations.
Label both first-box roles clearly as two modules, or show six separate boxes and preserve the independent safety gate.
Prefer “would,” “is designed to,” and “proposes” when implementation or testing has not occurred; keep deployment caveats.
What is particularly strong
Engineering relevance. It clearly links UPS/MV equipment, electrical capacity, cooling headroom and commissioning evidence without claiming a completed autonomous deployment.
TEAM A | INTERNAL TECHNICAL & EDITORIAL REVIEW | NOT FOR PUBLICATION Review appendix p. 1
Safety framing. Independent approval, fail-closed behavior, qualified human authority, provenance and conservative automation boundaries are all represented well.
Unusual scientific candor. The article explicitly says that a deterministic controller tied the coordinated arm under matched inputs, and that the remaining digital-twin evaluation is still planned. This avoids an unjustified “AI is superior” claim.
PAGE-BY-PAGE EDITS — EXACT ACTIONS
Pages 1 and 6 — professional credentials
Current: “Raja Kumara Swamy Donthula, P.Eng., Senior Member, IEEE …” (photo caption, p. 1; author biography, p. 6).
Replace until licence issuance: “Raja Kumara Swamy Donthula, Senior Member, IEEE …”
Reason: Being academically assessed, an applicant, an EIT registrant, or awaiting a competency decision is not the same as holding an issued P.Eng. licence. If approval is subsequently granted, confirm the official registration status before reinstating the designation. A future expected approval does not validate the title on an earlier publication date.
Pages 2–3 — six agent roles versus five visual blocks
The text identifies equipment prognostics, topology/capacity, risk/diagnosis, response orchestration, independent safety verification and operator support. The diagram combines the first two into “Prognostics + Topology/capacity” but does not explain that this is a compact presentation of six roles in five boxes.
Suggested figure subtitle: “Six logical roles are represented by five grouped boxes; prognostics and topology/capacity remain separate agent functions.” Better: split the first box into two labeled boxes and retain the visible independent gate and audit ledger.
Pages 2–4 — distinguish design intentions from deployed facts
The article correctly introduces a proposal, but sentences such as “All of them write to a shared, append-only record” and “It checks each candidate” can be mistaken for a deployed software implementation.
Suggested replacements: “In the proposed architecture, each role would record its inputs and recommendations in a shared logical audit ledger.” “The separate safety-gate role is designed to check candidate actions against facility-approved constraints and withhold clearance when required information cannot be verified.”
Do not imply that the shared ledger is tamper-proof or cryptographically protected; the manuscript describes a logical audit requirement, not a proven secure log. Also preserve that arc-flash calculations come from approved external studies rather than from the framework itself.
Pages 4–5 — results and evaluation maturity
The shared-bus case is accurately discussed: the coordinated arm enforces a cooling floor and prevents capacity/floor breaches, but an equally informed deterministic real-time rule matched it. Keep this negative result. It is one of the most defensible technical details in the feature.
The paper reports ten illustrative pilot experiments in two five-experiment suites, not just five in total. The article’s “five small pilot simulations” wording is defensible if referring only to the first suite, but a short clarification would prevent a reader from thinking the manuscript had only five illustrative studies.
Suggested sentence: “The manuscript reports two sets of five small illustrative pilot tests, separate from the fuller physics-informed digital-twin validation that remains future work.”
On page 5, where the future full evaluation is described, change “toolchain, solver settings, topology, ratings and random seeds are frozen before any comparative run” to “the proposed protocol calls for these to be frozen before the full comparative evaluation.” This avoids implying the specified full study has already been completed.
TEAM A | INTERNAL TECHNICAL & EDITORIAL REVIEW | NOT FOR PUBLICATION Review appendix p. 2
Page 6 — closing attribution
Suggested complete closing: “The framework discussed here is described in his research manuscript, ‘Uncertainty- and Risk-Aware Multi-Agent Decision-Support Framework for Safety-Critical Data-Center Electrical Operations.’ The concept and early pilots have not been validated in a live facility.”
PAGE-BY-PAGE EDITS — EXACT ACTIONS
Pages 1 and 6 — professional credentials
Current: “Raja Kumara Swamy Donthula, P.Eng., Senior Member, IEEE …” (photo caption, p. 1; author biography, p. 6).
Replace until licence issuance: “Raja Kumara Swamy Donthula, Senior Member, IEEE …”
Reason: Being academically assessed, an applicant, an EIT registrant, or awaiting a competency decision is not the same as holding an issued P.Eng. licence. If approval is subsequently granted, confirm the official registration status before reinstating the designation. A future expected approval does not validate the title on an earlier publication date.
Pages 2–3 — six agent roles versus five visual blocks
The text identifies equipment prognostics, topology/capacity, risk/diagnosis, response orchestration, independent safety verification and operator support. The diagram combines the first two into “Prognostics + Topology/capacity” but does not explain that this is a compact presentation of six roles in five boxes.
Suggested figure subtitle: “Six logical roles are represented by five grouped boxes; prognostics and topology/capacity remain separate agent functions.” Better: split the first box into two labeled boxes and retain the visible independent gate and audit ledger.
Pages 2–4 — distinguish design intentions from deployed facts
The article correctly introduces a proposal, but sentences such as “All of them write to a shared, append-only record” and “It checks each candidate” can be mistaken for a deployed software implementation.
Suggested replacements: “In the proposed architecture, each role would record its inputs and recommendations in a shared logical audit ledger.” “The separate safety-gate role is designed to check candidate actions against facility-approved constraints and withhold clearance when required information cannot be verified.”
Do not imply that the shared ledger is tamper-proof or cryptographically protected; the manuscript describes a logical audit requirement, not a proven secure log. Also preserve that arc-flash calculations come from approved external studies rather than from the framework itself.
Pages 4–5 — results and evaluation maturity
The shared-bus case is accurately discussed: the coordinated arm enforces a cooling floor and prevents capacity/floor breaches, but an equally informed deterministic real-time rule matched it. Keep this negative result. It is one of the most defensible technical details in the feature.
The paper reports ten illustrative pilot experiments in two five-experiment suites, not just five in total. The article’s “five small pilot simulations” wording is defensible if referring only to the first suite, but a short clarification would prevent a reader from thinking the manuscript had only five illustrative studies.
Suggested sentence: “The manuscript reports two sets of five small illustrative pilot tests, separate from the fuller physics-informed digital-twin validation that remains future work.”
On page 5, where the future full evaluation is described, change “toolchain, solver settings, topology, ratings and random seeds are frozen before any comparative run” to “the proposed protocol calls for these to be frozen before the full comparative evaluation.” This avoids implying the specified full study has already been completed.
TEAM A | INTERNAL TECHNICAL & EDITORIAL REVIEW | NOT FOR PUBLICATION Review appendix p. 2
Page 6 — closing attribution
Suggested complete closing: “The framework discussed here is described in his research manuscript, ‘Uncertainty- and Risk-Aware Multi-Agent Decision-Support Framework for Safety-Critical Data-Center Electrical Operations.’ The concept and early pilots have not been validated in a live facility.”
PUBLICATION QUALITY, EB-1A EVIDENCE & RELEASE GATE
Journalistic quality and source attribution
Author, date, URL. Before the final outlet publication, record the byline, publication date, outlet identity, category, stable URL and any later editorial changes. The current PDF alone does not contain these metadata. Preserve the editorial correspondence.
Photograph and diagram. Confirm that the photographed face is authentic, the image may legally be published, the credit line is authorized, and digital overlays are described as illustrative. The figure must remain labeled as conceptual, not an image of a deployed control room system.
No publication-status inflation. Refer to the underlying work as a manuscript or research study unless journal/conference acceptance is confirmed. Do not describe pilot experiments as independently validated hardware results or the content as a peer-reviewed journal article.
Headline. The current headline accurately signals decision support rather than autonomous switching. Retain the explicit “human-supervised” qualifier; shorten only if requested by the editor.
EB-1A documentation — separate from media approval
The article discusses the named researcher’s actual engineering work at substantive length, which can be useful. However, publication by TechAnnouncer does not automatically satisfy the EB-1A “published material about the person” criterion. An attorney must assess the outlet’s professional audience/standing and comparative readership; TechAnnouncer openly offers press-release and paid promotional services. Preserve the terms of placement and disclose any payment or team-arranged submission to counsel rather than presenting it as independently solicited journalism.
For evidence, obtain and preserve: (1) published page/PDF with headline, journalist name, publication date and URL; (2) evidence of outlet audience/readership and editorial process; (3) editorial revision/acceptance correspondence; (4) original commissioned image licence; and (5) the relevant research manuscript and public status. Avoid claiming editorial acceptance proves external scientific verification.
Patent disclosure check
If any part of the technical architecture is intended for the contemplated German patent filing, have a patent attorney compare what this article discloses against the invention description before publication. Media disclosure may affect patent novelty and the timing of priority filings. This note is a prompt for legal clearance, not an assertion that the article necessarily destroys patentability.
Release decision
CONDITIONAL GREEN SIGNAL: Team A can send a corrected version to the editor after the required factual and editorial edits. I would not approve this exact six-page PDF for publication with an unissued P.Eng. designation or an unfinished final sentence. Do not append these review pages to the version supplied for public publication; the appendix is an internal instruction sheet.
Source scope
Original: “Raja_Donthula_TechAnnouncer_Feature 2.pdf,” six pages, provided 24 Oct 2026. Compared with the eight-page working manuscript found in the research portfolio. Publisher website policies and general credential/immigration rules were consulted for contextual recommendations; the manuscript itself has not been independently replicated or verified against raw experimental data.
TEAM A | INTERNAL TECHNICAL & EDITORIAL REVIEW | NOT FOR PUBLICATION Review appendix p. 3
COPY-READY RESPONSE TO TEAM A
Hi Team A,
Thank you for preparing the fourth media feature for TechAnnouncer. I have completed the technical and editorial review, including comparison with the underlying research manuscript. The article is well structured and presents the engineering safety and human-supervision concepts accurately. I support moving toward publication without a major rewrite.
Before finalizing, please make these corrections:
- Remove “P.Eng.” from the opening photo caption and closing biography until my formal EGBC licence is granted. Please keep “Senior Member, IEEE.”
- Fix the missing closing quotation/punctuation around the manuscript title on the last page and the excessive blank space.
- Update the architecture diagram to clearly distinguish all six logical agent roles, or explain that the first two are combined into one visual block.
- Revise any operational wording that sounds like a deployed system: the architecture is proposed and the completed results are illustrative simulations. The full digital-twin validation remains future work.
- Clarify that the manuscript contains two sets of five pilot experiments, keep the honest comparison with the deterministic controller, and phrase the planned test-freezing procedure in future tense.
- Confirm the article byline, outlet publication date, final link, and photo rights/credits for the published version.
Please also confirm that the final wording has been checked against any pending patent filing to avoid premature disclosure of an invention. Once these points are corrected, please share the clean, publication-ready PDF for final approval. I do not recommend additional major changes to the topic or technical content.
Thank you.
Final verdict for Team A: Strong media feature; proceed after targeted corrections. Do not present this review appendix as part of the published article.
