AI Services · EP&S · Rio Tinto

Enterprise AI, delivered safely

The AI delivery and advisory team within EP&S, turning Rio Tinto's AI ideas into working, governed services.

See how it works or talk to us directly
01

Document AI

Contract, policy & form workflows

Finance, Procurement, and Commercial teams extract structured data, surface non-standard clauses, and flag risks. Clear review checkpoints and evidence on accuracy before anything goes to production.

Best for: document-heavy workflows where structure and human review matter

02

Knowledge / RAG

Standards, policy & knowledge Q&A

Safety, Technical Standards, and HR teams get grounded, cited answers over internal policies and knowledge bases. Controlled data access, citation expectations, and no shadow query paths.

Best for: governed internal knowledge where accuracy and traceability are non-negotiable

03

Workflow Automation

Repeatable process acceleration

Known, repeatable steps where AI adds classification, extraction, summarisation, or decision support. Often higher value than a generic chatbot, and always with a clear audit trail and named support owner.

Best for: structured, measurable workflows with clear controls and run ownership

04

Agentic

Multi-step bounded decisioning

Used selectively for multi-step work requiring tool use, context retrieval, and bounded decisioning. Autonomy limits, observability, and stop conditions are defined before build starts, not after.

Best for: complex orchestration where simpler patterns won't reliably solve the problem

What we do

Our service areas

We are accountable end-to-end: from qualifying demand to delivering governed, supportable AI services that enterprise functions can actually rely on.

Intake & Triage

One enterprise-facing front door for AI demand. We qualify fit, shape problem statements, and decide whether a request belongs with AI Services or another team.

AI Solution Delivery

Discovery, design, build, pilot, and production delivery of approved AI use cases, using LLMs, agentic patterns, RAG, document AI, and purpose-built models.

AI Advisory

Guidance on GenAI tools, build-vs-configure-vs-buy choices, safe-use patterns, vendor fit, and enterprise-ready architecture, before you commit to a build.

Delivery Standards

Evaluation approach, human-in-the-loop design, supportability expectations, release gates, and observability requirements, so AI services are production-grade from the start.

Run & Support

Run readiness, support-path design, monitoring, and improvement backlog discipline once a service goes live, so nothing becomes orphaned technical debt.

Reusable Patterns

Templates, reference flows, and a service catalogue of delivery patterns, so good solutions benefit more than one team and the whole enterprise learns faster.

Learn

New to AI? Start here.

Three short modules that cover the essentials: what AI is, how Rio Tinto is using it, and how to start a project the right way.

Module 1 thumbnail
01

AI Introduction

What AI is, how it works, and why it matters for enterprise teams.

Module 2 thumbnail
02

AI Within Rio Tinto

How Rio Tinto is approaching AI, what's in scope, and where the guardrails are.

Module 3 thumbnail
03

Starting AI Projects

The right way to scope, sponsor, and start an AI use case at Rio Tinto.

How we operate

Three principles that guide every decision

I

Safe

AI Services delivers within enterprise guardrails. Solutions must have appropriate governance, data controls, evaluation, observability, and support readiness before becoming production services.

II

Pragmatic

We prioritise measurable workflow improvement over technical novelty. We use the simplest delivery pattern that can reliably solve the business problem, avoiding AI theatre and unsupported experimentation.

III

Measurable

Every material engagement defines a measurable value hypothesis and success metrics before delivery begins. Live services track adoption, quality, cost, and business value, so we can make informed continue, scale, or stop decisions.

The problem

Sound familiar?

AI capability is appearing everywhere at once. Without a clear service model, demand scatters, ownership gets muddy, and orphan pilots turn into support debt.

Challenge

  • Scattered, unshaped AI demand
  • Vendor AI moving fast inside enterprise platforms
  • Functions want speed but readiness is uneven
  • Orphan pilots create support debt
  • Wrong technical pattern chosen for the job
  • AI activity scattered across pockets with no coherence

Our response

One enterprise-facing intake and triage path, with a named sponsor, problem statement, and success measure confirmed before any build starts.

Our process

Delivery gates: not just good intentions

Every engagement passes through explicit stages. No pilot becomes a service without a named sponsor, evaluation approach, support owner, and production gate sign-off.

1

Intake

Named sponsor, workflow owner, value hypothesis, data owner, sensitivity classification, and support owner confirmed before we invest discovery effort.

2

Discovery & Feasibility

We find the lightest delivery pattern that works, map required partners, and define the first evaluation approach and risk controls before any build starts.

3

Pilot Gate

Gate

Scope bounded, human review explicit, cost visible, rollback possible, support path named. Proceed, defer, reshape, redirect, or stop, decided explicitly.

4

Production Gate

Gate

Permissions, observability, runbook, escalation path, cost view, evaluation set, and handover ownership confirmed. Named service owner required before release.

5

Run & Improve

Adoption, quality, cost, incident, and value signals monitored continuously. Every live service has an explicit continue, improve, scale, or stop decision point.

How to work with us

Six ways we engage

We have an explicit vocabulary for how we work. Not every request becomes a build, and we can redirect or stop weak work cleanly.

Build modes

Deliver

Sponsor-backed use case

Discovery, build, pilot, and production delivery for a named enterprise use case with a real sponsor and measurable outcome.

Advise

Choosing tools or patterns

Decision framing, architecture guidance, build-vs-configure recommendation, and risk assessment when a team needs direction before committing.

Shape

Standards before scaling

Templates, delivery gates, evaluation approach, and service-catalogue entries when a capability needs common standards before it scales.

Support & exit modes

Operate

Live service ownership

Runbooks, incident paths, release management, and service improvement for AI tools or production services where we hold explicit operational remit.

Redirect

Wrong team or not ready

A fit decision, partner handoff, and re-entry conditions when a request belongs with another team or needs sponsor shaping before we invest effort.

Stop

Weak on value or fit

A clear stop decision with rationale and learning carried forward, when sponsor pull, value, data readiness, risk fit, or supportability is insufficient.

The team

Who we are

A small, focused delivery and advisory function, based across Brisbane and Perth, operating inside Enterprise Platforms & Services.

Alexey Molodchikov

Alexey Molodchikov

Manager, AI Services

Brisbane

Maria Ramiro

Maria Ramiro

Senior Specialist, Product Analyst

Brisbane

Awal Singh

Awal Singh

Senior Specialist, Data Science

Brisbane

Jocelyn Chai

Jocelyn Chai

Senior Analyst, Data Science

Brisbane

Alice Nguyen

Alice Nguyen

Specialist, Data Engineering

Brisbane

Ethan Phan

Ethan Phan

Specialist, Data Engineering

Brisbane

Tsolmon Otgonbold

Tsolmon Otgonbold

Senior Specialist, Data Engineering

Perth

Product & BA
Data Science
Data Engineering

Have an AI challenge worth solving?

Start with a named sponsor and a problem statement. We'll qualify the fit, shape the approach, and tell you honestly whether it's ready to build, or not yet.

1. Reach out to Maria Ramiro. She owns intake and will shape the problem statement with you
2. Come with a named sponsor and a rough workflow in mind, not a solution
3. We'll tell you honestly if it's ready to build, needs more shaping, or belongs with another team
Talk to us aiservices@riotinto.com