The Challenge
The Questions a Maintenance Team Asks Every Day
At TAQA in Oman, Qura looked after rotating equipment first as a Maintenance Engineer, then as Workshop Supervisor — representing rotary equipment in HAZOP meetings with clients including PDO, Shell, and Oxy. The equipment was not the hardest part of the job. The hardest part was finding the answer to a simple question when the person who knew it was on a rotation break, or the certificate that proved it was in a different system from the job card that explained it.
Oil and gas operators and field-services companies run large fleets of rotating equipment and pressure vessels. Each asset carries a long tail of records: its place in the equipment hierarchy, its job cards, its inspection and calibration certificates, its spare parts, and the people certified to work on it. In most organisations those records live in different places — an ERP module, a spreadsheet of certificate expiry dates, a shared drive of scanned test reports, and the memory of the engineers who have maintained the equipment for years.
The consequence is that the questions reliability engineers actually need answered are slow and expensive to answer. Why did this pump fail twice in a quarter? Which certificates on this package expire before mobilisation? What is the real MTBF of this equipment class, and which failure modes drive its maintenance cost? Answering any of them means reconciling records by hand — linking a certificate to a job card to a failure trend for the same asset — and the answer is only as good as the engineer doing the reconciling.
Generic CMMS products treat this as a data-entry problem. Generic AI assistants treat it as a search problem, and produce fluent answers that cannot be traced to a single record. Neither reflects how an experienced maintenance and reliability engineer actually reasons about an asset — and in a regulated, safety-critical industry, an answer that cannot be traced is an answer that cannot be used.
The Approach
Building the Product Around Qura's Practice
Starting From the Practitioner, Not the Data Model
RelyentX was built with Qura as domain expert and co-author. His career runs through every layer of the problem: maintaining pumps, compressors, and piping at Petrogulf in Egypt; designing and installing pumping systems and overhauling centrifugal pumps in Oman; and six years at TAQA, where he cut equipment downtime from three months to one week, ran the annual optimisation of the company's CMMS, planned work in SAP Plant Maintenance, refined rotary maintenance procedures to reduce failure rates by 5%, and commissioned ten multi-stage pumps through site acceptance testing. He now works on projects and proposals at National Oilwell Varco.
In parallel, he completed an MSc in Artificial Intelligence at the University of Essex and is studying for an MS in Computer Science at Georgia Tech, with a focus on where AI genuinely helps in predictive maintenance and asset reliability. His published research — on reducing human error in maintenance processes using multi-agent systems, and on the Artificial Maintenance Management Agent (AMMA) framework for predictive maintenance automation — is the direct precursor to RelyentX's copilot. He holds CMRP, ASQ CRE, and PMP certifications. RelyentX is where those two tracks meet: a decade of field judgment, structured by someone who also understands the AI.
An Equipment Taxonomy That Mirrors Engineering Practice
The first decision was how an asset should be described. Rather than a flat equipment list, RelyentX uses a five-level taxonomy — Family, Class, Type, Model, Sub Type — with class-specific properties, parent/child composition, and lifecycle status. This is how reliability engineers already think about a fleet: failure modes, inspection requirements, and spare-part strategies attach to a class or model, not to an arbitrary tag number. FMEA is captured at model level for the same reason, so every unit of that model inherits the failure-mode analysis behind it.
Certification and Maintenance as Engineering Evidence
Inspection and certification were modelled on the disciplines Qura scheduled and signed off at TAQA — hydro test, load test, MPI, UTI, visual inspection, SRV calibration, and thickness measurement — as a closed set rather than free text. Calibration certificates are generated in the application from readings taken against a traceable test device. Job cards carry one of three work types, a system-stamped close date, and cost inputs, so that MTTR and maintenance cost are derived from the record rather than typed in after the fact.
Reliability Metrics Rebuilt From Events
MTBF, MTTR, MTTF, availability, backlog, overdue work, cost Pareto, and failure trends are rebuilt from the underlying domain events rather than stored as summary figures. The engineer reviewing a reliability number can always get back to the job cards and failures that produced it — the same traceability a senior engineer would demand before acting on a metric.
An AI Copilot That Cites Its Sources
The AI layer was designed around Qura's original question — a maintenance assistant available 24/7 — and a single principle: an answer about a safety-critical asset is only useful if the engineer can see where it came from. The RelyentX copilot searches each tenant's own vector index and equipment knowledge graph, and shows the source records alongside every answer. It is read-only by default. Write mode is opt-in and resets each session; when enabled, the copilot can only propose actions from a server-side allow-list, nothing runs until a user approves that specific action, and approved actions execute through the same authorised path as the dashboard. Answers and exports are treated as drafts for engineering review, not as decisions.
The Product
One System of Record for Equipment, Certification, and Reliability
RelyentX is an AI-native enterprise asset management (EAM) and CMMS platform for oil and gas. It brings the asset registry, maintenance and work orders, inspection and certification, reliability analytics, inventory and spare parts, and projects, people, and suppliers into a single multi-tenant system of record — with no ERP required.
Inventory is managed with valuation, usage classification, and EOQ and reorder points calculated with Wilson's formulae, with an approval gate on every issue. Projects and packages support equipment movement requests, while the employee directory tracks certifications and fit-to-work status alongside a supplier register. The platform is fully bilingual in English and Arabic, with complete right-to-left layout for operators across the Gulf region.
Security was designed for the industry it serves. Each tenant has its own identity organisation, vector collection, and graph database, with row-level security protecting pooled data. Compute, storage, databases, and outbound email are pinned to EU regions, and the AI services that sit outside that boundary are disclosed openly rather than buried.
The result is a tool that answers the questions a reliability engineer asks — why did this pump fail twice in a quarter, which certificates expire before mobilisation, where is maintenance cost concentrated — with answers traceable to the job cards, certificates, and events behind them.
10
Years of Qura's maintenance and reliability practice in oil and gas, built into the product's design
5
Levels of equipment taxonomy — Family, Class, Type, Model, Sub Type — mirroring how engineers structure a fleet
7
Inspection and certification disciplines modelled natively, from hydro test to SRV calibration
"What if you had a maintenance assistant who works for you 24/7 and knows the answer to every question about your equipment?"— Qura, Maintenance & Reliability Engineer
A decade of reliability knowledge, available around the clock.
If your operation runs on expertise that generic software cannot capture, Praxa builds the product that makes it operational.
Explore RelyentX