Integrity engineers and chief inspectors spend a large share of their week answering questions that already have answers somewhere in the record: which circuits in Crude Unit 2 have a short-term corrosion rate above 15 mpy, which vessels have open API 579 temporary repairs due at the next outage, or which clause of API 510 governs a specific half-life interval calculation.
Generic AI chatbots cannot answer these questions safely because they do not know the plant's asset hierarchy and will guess at code clauses or invent numbers when uncertain.
Current state before intelliSPEC
- Hours spent cross-filtering spreadsheets: Answering a plant manager's or auditor's question about overdue CMLs or recurring heat-exchanger tube leaks required exporting multiple tables and building pivot charts by hand.
- Distrust of ungrounded AI: Corporate security and engineering teams rightly blocked public LLM tools that could not cite the exact internal procedure or API standard paragraph behind a recommendation.
- Risk of unapproved record edits: Any automation that could alter an inspection interval, risk score, or workflow state without an engineer's signature was unacceptable in a PSM-regulated facility.
Configured path
The engineering team enabled Spectra across its intelliVISION, intelliINSPECT, intelliNDT, intelliINTEGRITY, and intelliCOMPLY workspace, paired with voice-to-form in intelliEDGE.
- Tenant-scoped retrieval with exact citations: Spectra queries only the tenant's live asset graph and uploaded procedures/standards library, returning the document name, page, and paragraph alongside every answer.
- Gloved-hand voice-to-form in the field: Inspectors dictate visual findings into
; Spectra maps the transcript into structured form fields while keeping the inspector in control of final review and sign-off. - Human approval gates on every mutation: All 59 agentic actions in Spectra are classified by risk; read queries execute immediately, while any create or update prepares a draft and waits for an authorized user with the required RBAC permission to approve it.
Measured outcome
- Instant cited audit answers: Engineers ask plain-language questions across thousands of assets and receive verifiable tables linked to the underlying inspection records and code clauses.
- Continuous register hygiene: Spectra flags duplicate asset naming, missing damage-mechanism assignments, and expiring inspector credentials before they surface in an audit.
- Zero autonomous mutations: Every calculation and record change retains a human author and approver in the hash-chained audit trail.