Product tour

Five stages from evidence to signed work package

This is the GridAPM workbench walkthrough on synthetic data: evidence intake, the quality gate, agent reasoning, the engineer's sign-off, and the work package that ships. Every screen below is the designed product UI — including one you can try.

  1. Intake
  2. Correlate & quality gate
  3. Agent reasoning
  4. Engineer sign-off
  5. Work package & report

AgenticGrid Pro Transformer APM

The other GridAPM product ProtectionAI Relay testing

Compare both products

The walkthrough

One transformer, end to end

Follow synthetic asset TX-47 · 230/69 kV through the five pipeline stages. The rail beside the screens tracks where you are.

Intake

Approved evidence streams — DGA, partial discharge, SFRA, thermal, inspection records — are read from your systems.

Every record keeps its source, timestamp, and provenance.

A pilot starts with files your team already has: lab DGA exports, PRPD and SFRA test sets, IR imagery, inspection notes. The workbench parses each one and pins its provenance — source system, timestamp, and owner — to the asset record.

GridAPM Workbench — local deployment Synthetic demo data
TX-47 230/69 kV · 60 MVA · 1998 · ONAF H2 trend 148 ppm
  • DGA Passed TX-47_DGA_2026-06-30.csv Oil lab LIMS export
  • PRPD Passed TX-47_PRPD_HVB_2026-06-12.pdx PD monitor export
  • SFRA Passed TX-47_SFRA_2026-05-18.sfra vs 2019 baseline
  • Thermal Corrected — clock drift +42 min TX-47_IR_2026-06-28.rjpg IR camera
  • TX-47_nameplate_scan.pdf Parsed
  • TX-47_loadprofile_2026-Q2.csv Normalizing…
  • TX-47_oil-quality_2026-06-30.csv Queued
  • Local-first
  • No OT connection
  • Audit log on
Product screen: evidence intake for transformer TX-47 — DGA, PRPD, SFRA and thermal source files passing the quality gate before analysis.

What the engineer sees The TX-47 asset header, one provenance chip per source file, and the import queue doing its work.

Correlate & quality gate

Signals are aligned per asset and screened for gaps, unit errors, and stale data before any reasoning starts.

Rejected inputs are logged with the reason they failed the gate.

Before any AI touches the data, the quality gate aligns signals per asset and screens for gaps, unit errors, and stale records. Corrections are applied in the open — here, an IR camera clock drift — and rejections are logged with their reason.

GridAPM Workbench — local deployment Synthetic demo data
TX-47 230/69 kV · 60 MVA · 1998 · ONAF H2 trend 148 ppm
  • DGA Passed TX-47_DGA_2026-06-30.csv Oil lab LIMS export
  • PRPD Passed TX-47_PRPD_HVB_2026-06-12.pdx PD monitor export
  • SFRA Passed TX-47_SFRA_2026-05-18.sfra vs 2019 baseline
  • Thermal Corrected — clock drift +42 min TX-47_IR_2026-06-28.rjpg IR camera
  • TX-47_nameplate_scan.pdf Parsed
  • TX-47_loadprofile_2026-Q2.csv Normalizing…
  • TX-47_oil-quality_2026-06-30.csv Queued
  • Local-first
  • No OT connection
  • Audit log on
Product screen: evidence intake for transformer TX-47 — DGA, PRPD, SFRA and thermal source files passing the quality gate before analysis.

What the engineer sees Gate marks on every source chip: three passed, one corrected, nothing silently accepted.

Agent reasoning

Agents draft a condition assessment, citing each piece of evidence and flagging contradictions instead of hiding them.

Every draft carries its evidence chain and a confidence band.

Agents draft a condition assessment with the evidence chain attached: cited gas values, ratio analysis, and — deliberately — the one signal that disagrees. Contradictions are flagged, never smoothed over, and the draft carries a confidence band.

GridAPM Workbench — local deployment Synthetic demo data

AI draft — awaiting engineer review

  • DGA — thermal signature

    C2H4 58 ppm exceeds the 50 ppm screening limit; CH4 74 ppm. Ratio C2H4/C2H6 = 2.1 → thermal fault 300–700 °C (IEC 60599, T2). Duval Triangle 1 places the 2026-06-30 sample in T2, close to the DT boundary.

  • Thermal — IR scan 2026-06-28

    Hot spot 98 °C at the B-phase LV connection, +23 K against the A-phase reference at comparable load (0.82 pu); top-oil 71 °C.

  • SFRA — no bulk winding movement

    2026-05-18 sweep vs 2019 baseline: correlation 0.99 below 2 kHz — no indication of core or bulk winding displacement. The 10–200 kHz band deviates (correlation 0.87), consistent with a local connection change rather than winding movement.

  • Contradiction flagged

    H2 rise also fits partial discharge

    H2 148 ppm (limit 80) and C2H2 11 ppm also fit a discharge; PRPD 2026-06-12 shows an emerging pattern at 30°–110° (peak 460 pC). The draft keeps the thermal fault as primary; repeat DGA in 30 d and keep the PD monitor trending to separate the two hypotheses.

0.78 — Moderate-high agent/dga-v2.3

Re-sample oil in 30 days; keep the PD monitor trending on the HV-B bushing; thermographic follow-up on the B-phase LV connection; inspect and re-torque at the next planned outage.

  • Local-first
  • No OT connection
  • Audit log on
Product screen: the agent's chain of evidence for TX-47 with cited gas values, one flagged contradiction, a confidence band, and the persistent AI-draft ribbon awaiting engineer review.

What the engineer sees The chain of evidence, a flagged contradiction, the confidence band, and the persistent AI-draft ribbon.

EngineerHuman gate

Engineer sign-off

A qualified engineer approves, edits, rejects, or escalates the draft. Nothing ships without a named reviewer.

The decision, the reviewer, and the timestamp enter the audit log.

Stage 4 is the human gate. A named engineer approves, edits, rejects, or escalates the draft — and each decision stamps the append-only audit log. Nothing becomes a recommendation without this signature.

Try it — approve or reject the draft

GridAPM Workbench — local deployment Synthetic demo data

AI draft — awaiting engineer review

Reviewer: M. Okafor, P.E. · Senior substation engineer

Re-sample oil in 30 days; keep the PD monitor trending on the HV-B bushing; thermographic follow-up on the B-phase LV connection; inspect and re-torque at the next planned outage.

Awaiting engineer decisionNothing ships without a named engineer's decision.

  1. intake · 7 files received, 4 sources correlated
  2. quality gate · passed — 1 correction (IR clock drift)
  3. agent/dga-v2.3 · draft v2 created — confidence 0.78
  • Local-first
  • No OT connection
  • Audit log on
Product screen: engineer sign-off for TX-47 — Approve, Edit, Reject or Escalate; each decision stamps the audit log. Try the buttons: the states are pre-rendered.

What the engineer sees The four decision actions, the reviewer's identity, and the audit log updating with each choice.

Work package & report

Approved decisions become maintenance work packages and audit-ready reports for your CMMS and stakeholders.

Each report traces back to the raw records it was built from.

Approval turns the draft into a work package: a prioritized task list with an outage-window note, ready for CMMS handoff and export as an audit-ready evidence pack.

GridAPM Workbench — local deployment Synthetic demo data
WP-2026-0141 TX-47 thermal follow-up APPROVED

Signed: M. Okafor, P.E. · 2026-07-10 14:32

  1. P1 Oil re-sample + DGA lab analysis due 2026-07-24 · Field services
  2. P2 IR follow-up scan, B-phase LV bushing due 2026-08-07 · Thermography
  3. P3 Inspect & re-torque LV connection outage window 2026-09 · Maintenance
CMMS handoff SAP PM work order draft · SS12-TX-47 Ready
  • PDF report
  • CSV task list
  • CMMS XML
  • Local-first
  • No OT connection
  • Audit log on
Product screen: work package WP-2026-0141 for TX-47 — prioritized task list, CMMS handoff and export options after engineer approval.

What the engineer sees Work package WP-2026-0141 with its task list, handoff status, and export options.

FAQ

Product tour boundaries

What role does GridAPM play in operations?

GridAPM is decision support: a local-first workbench for controlled transformer APM pilots. Engineers hold final authority — it performs no autonomous control and claims no final diagnostic authority.

Can a pilot start offline?

Yes. Local-first evaluation on approved historical evidence is the default starting point. OpenAI access and any live integrations are separately reviewed and controlled.

What is the main output of a pilot?

A pilot produces evidence packs, reviewer notes, missing-evidence lists, AI draft rationale, and human-approved work-package context — the same artifacts shown in this tour.

Run this walkthrough on your own fleet

A controlled pilot follows these same five stages on your approved records, with your engineers at the gate.

Type to search research, platform pages, and tools.