Equipping a world-class maintenance engineering team with the AI tools and mental models to lead from the front

THE PROJECT

Secora’s brief was to xxxx xxxxxx xxxxxxx xxxxxxx xxxxxxxx xxxxxxxxx xxxx xxxxxxx x xxxx.

WHO WE’RE WORKING WITH

BHP's Maintenance Engineering Centre of Excellence (MECoE), Workstream 2 — a specialist team applying data science and asset management expertise across BHP's operations.

THE CHALLENGE

Maintenance datasets at BHP's scale contain vast stores of operational intelligence — embedded in work order descriptions, notification long text, and system event records. The problem is structural: that intelligence is largely inaccessible at scale through conventional keyword search or manual review. The team had the domain expertise and the data. They needed the AI capability to unlock what was already there.

WHAT SECORA DID

Across Q3 FY26 (January–March 2026), Secora delivered a structured AI investigation program for MECoE under Workstream 2, working within BHP's approved AI toolchain — Amazon Bedrock, SageMaker, and GitLab.

Three AI capabilities were investigated and demonstrated. The first was LLM-based long-text analysis: using Claude via Amazon Bedrock to read and classify hundreds of Notification and Work Order descriptions simultaneously — surfacing failure modes, material delay causes, and deferred work patterns at a scale previously accessible only through manual, one-by-one review. The second was AI coding assistance: demonstrating that Claude Code could write, refactor, and debug Python Notebooks within SageMaker, eliminating code authorship as a bottleneck and opening the prospect of sophisticated automated analytical workflows. The third was specialised agent development: confirming that MECOE analysis routines can be codified into supervised agents that operate continuously alongside analysts — codifying lines of enquiry and freeing the team to pursue the next problem rather than repeat the last one.

Alongside the investigations, Secora delivered a three-session AI upskilling program covering AI fundamentals, agentic workflow architecture, and AI cost management. Participants identified concrete prototype candidates organically across the sessions — a signal that the domain expertise to drive this capability was already present in the team.

THE RESULT

The engagement closed with a clear FY27 roadmap: LLM-assisted notification long-text analysis as the first production use case, Claude Code embedded as standard practice in Notebook development, and at least one supervised agent workflow progressed to prototype. All code lives in the team's own GitLab repositories. All skills and vocabulary belong to the team. Secora's stated aim — to leave capability behind, not dependency — was the design principle for the entire engagement.