A nationwide MEP contractor runs thousands of jobs at once. The information that fixes a Tuesday-morning trip on a six-month-old commissioning lives somewhere — in the handover binder, the subcontractor email thread, the riser diagram, the punchlist. The problem is not that it does not exist. The problem is that it is not findable when it matters.
What this industry actually runs on
MEP — electrical, HVAC, fire, plumbing — runs on project records, code references and commissioning evidence. Every project has its own handover binder, its own subcontractor list, its own punchlist, its own commissioning sign-off. Service teams operate against jurisdictional codes that differ by region and that update on cycles measured in years. The contractor’s senior service engineers carry the institutional memory; when one of them retires, three months of someone else’s first-call independence retires with them. The information substrate is not documents in a folder; it is documents organised by project, by jurisdiction, by subcontractor, by commissioning date.
Why this industry breaks generic AI
A generic RAG pipeline pointed at all your project docs gets the wrong answer the moment two projects share an equipment model. The HVAC unit installed at Project X in 2022 is not the same as the visually identical unit at Project Y in 2024 — different commissioning, different setpoints, different code revision, different subcontractor. Generic models have no mechanism to scope a code reference to the jurisdiction in front of the engineer; they will confidently quote a national standard for a job that runs under a regional code variant. Project-aware retrieval — scoping every query to the project, the jurisdiction, and the as-built record — is not a feature you bolt on after.
How Opero shows up here
- Project-aware knowledge. Ask about Project X and the agent knows the panel schedule, the riser diagrams, the commissioning record, the punchlist — scoped to that project before retrieval, not across the firm’s full archive. AI Workflows configured against the project register handles the scoping.
- Code reference grounded to the local jurisdiction. The agent retrieves the version of the standard that applies to this site, this contract, this revision date — not the most-cited public copy.
- Service handovers as the index, not a brain dump. Voice in, structured out. The handover from a commissioning engineer becomes a queryable record with the project, the equipment IDs and the open punchlist items tagged at ingest.
- Audit log on every retrieval. When the quality team asks six weeks later why a particular setpoint was changed on a particular date, the answer is in the log.
A real deployment
At one of the Nordics’ larger MEP contractors, the service organisation manages thousands of project records across multiple disciplines. The work the senior engineers used to do — being the recall mechanism for handover details — has shifted: the agent fields the routine recall questions, and the senior engineers spend their time on the cases where the retrieval surfaces a project-document gap. Technicians return a second time when the first answer earned their trust; that second visit is the adoption signal. Other MEP service organisations in the segment run similar patterns.
Where to look next
Three pages anchor the rest of the read: the workflow engine behind the project-aware retrieval, the operations-manager persona that runs on it, and a deployment where the corpus earned technicians’ second visit.
- AI Workflows — the engine for project-aware retrieval and handover-to-record automation.
- Operations & plant managers — the operational-hygiene persona.
- Nize Equipment — deployment shape; not MEP (Nize is large-format print), but the trust-accrual dynamics translate; a dedicated MEP case study is planned.