From thousands of documents to AI knowledge hub.

Queensland Rail

Sector: TRANSPORT
Stage: PROTOTYPE → PRODUCTION
Deliverables: GEN-AI SEARCH PORTAL, AI TRAINING AVATAR POC

We turned thousands of safety manuals into an Azure/OpenAI-powered knowledge hub, unlocking fast, trusted answers for 7,300+ rail workers.

INTRODUCTION

  
  Queensland Rail (QR) moves passengers and freight
  across 6,600 km of track and employs more than 7,300 people.
  Its Safety Environment Management System (SEMS) held
  decades of vital procedures,
  but staff spent hours hunting through PDFs.
  The brief: make SEMS conversational like ChatGPT, open it to partners,
  and drastically reduce time spent finding answers.

VISION

Put all rail-safety knowledge “on tap” for any employees, anytime.

We kicked off with a generative-AI proof-of-concept that showed semantic search could tame QR’s document mountain. Success green-lit a production roll-out to operationalise the app and drive organisation-wide adoption.

Discover & Design

Workshops That Mapped Risks and Proved AI’s Potential

We kicked off with a series of discovery workshops involving safety, engineering, and ICT teams to deeply understand the policy landscape and user needs. Together, we mapped end-to-end user journeys, catalogued critical data sources, and identified risk controls that would shape how the assistant answered queries.

To prove the concept early, we built and demoed a RAG-powered prototype on Microsoft Azure (powered by OpenAI models) that answered natural-language queries with source-linked citations — showing teams what was possible, fast.

A Rail-Safe Portal Built Securely Inside Azure

Hands-On Training

Once validated, we architected and delivered the Advanced SEMS Portal directly within QR’s Azure tenancy — ensuring compliance and ownership from day one. The system integrated blob storage, Cognitive Search, and OpenAI embeddings to power intelligent, context-aware search.

Early Adoption, Rapid Feedback, and Measurable Impact

Pilot & Scale

KPI dashboards were implemented to track usage, latency and answer accuracy — giving leadership clear visibility into performance gains as adoption grew.

We launched with 200 early adopters to validate usability and impact. Their feedback drove UI iterations and refinements while we automated nightly ingestion of over 4,000+ PDFs into the system.