Microsoft’s Azure OpenAI Service is being tapped to transform the management of technical document libraries at Main Roads Western Australia (Main Roads), the transport agency responsible for the state’s road network.

Through machine learning - specifically generative AI, thousands of documents in the public facing, web-hosted Main Roads’ Technical Library will soon be searchable with plain language prompts, and their relationships organised and dynamically mapped. 

Being developed in partnership with GHD Digital, the transformation arm of professional services firm GHD, the platform will enable new processes to maintain consistency and currency of the library’s documents, resulting in significant productivity gains and risk mitigation

The Main Roads Technical Library is a set of over 3,000 public-facing and inter-referenced policies, regulations, guidelines, standards and drawings, detailing how roads infrastructure is to be designed, developed, constructed and operated.

The library contains a wide variety of file types and document formats.

Custodianship of these documents is distributed across many branches of Main Roads, including Network Operations Planning. When one document is updated, the changes must be communicated and reflected in all referencing documents to ensure accuracy and consistency.

Currently, this process is time-consuming due to manual cross-referencing checks, and Network Operations Planning have been seeking an innovative alternative.  

“Multi-million-dollar contracts and projects are dependent on the precision of our Technical Library,” said Bita Charehjoo, Network Operations Planning Manager at Main Roads.

“Having spent many years developing and managing the policies, guidelines, specifications and technical drawings - and many hands-on hours checking the validity of edits and updates, there has to be a smarter, faster way of indexing and cataloguing documentation.”

Sarah Dods, Region Leader of Advanced Analytics and AI at GHD Digital, said: “This new generative AI approach seeks to automatically identify edits across impacted documents, draw inferences and communicate changes to necessary stakeholders, reducing a process that used to take days to complete to just minutes.”

The solution is being developed using large language models (LLMs) to read document and image text, identify references in various formats and extract relevant information much faster than a human.

Its function is to find references and governance metadata in both text documents and drawings, and to provide plain English search with document retrieval. 

Designed within Microsoft Azure, the solution integrates pre-trained LLMs and Azure AI Search, Microsoft’s AI-powered information retrieval platform, in a unique architecture. This setup allows Main Roads’ team members to input, read and respond in their everyday language. It also enables them to create new notification workflows that simplify the process of keeping information up to date and aligned.

Once notified, changes are manually reviewed by referring document custodians, to determine the extent to which the changes need to be reflected in their own updates.

“Given the purpose of the documents, to guide the construction, operation and maintenance of major infrastructure, we are expecting that document custodians will want to review the changes in person, once they are made aware there has been a change,” said Dods.  

"Consistency and reliability are so critical in this context and so it is exciting to see the innovative way in which Main Roads is exploring the use of AI to keep their essential documents in check," says Sarah Carney, Chief Technology Officer at Microsoft Australia and New Zealand.