‘ChatBDP’ tool transforms access to internal information and documentation, addressing limitations with traditional search methods
A custom AI engine trained to search BDP’s internal knowledge base and give instant answers to queries related to technical guidance, process documents, marketing and bid material is being rolled out across the business.
The ChatBDP tool was developed in collaboration with data science consultancy Simpson Associates, based on an in-house proof of concept built using Open AI’s ChatGPT chatbot. The system allows users to run queries against information using natural language processing to sift through its documents and generate concise summaries in response.
Many companies experience problems searching internal data due to the sheer quantity of documents and shortcomings of traditional software searches. BDP turned to AI as an alternative to traditional searches on the intranet and in SharePoint libraries, which were taking too long.
Alistair Kell, a principal and chief information officer at BDP, said: ‘People were finding it harder and harder to find relevant information on key topics within the business, doing keyword searches through file systems or intranets and being provided with multiple different information sources that needed to be read, analysed and understood for relevance.’
The proof of concept trained on around 200 BDP documents impressed the chief executive and other board members, who gave the green light to develop an enterprise-level AI engine, integrating ChatGPT into a custom web application housed in BDP’s existing Microsoft Azure tenant.
‘It's an internal version of ChatGPT looking only at our own knowledge,’ said Kell. ‘Staff run a plain language query and get a couple of A4 pages of summary, plus direct access to all the documentation initially assessed to create the summaries.’
Data on the system forms the basis of training for the AI model and includes all the studio’s design processes, project controls and guidance. In a bid to strengthen adoption of the ISO 9001 quality management system for Building Safety Act compliance, ChatBDP provides links to summaries of the top six non-conformities around project controls, ‘helping drive our continual improvement for ISO standards,’ said Kell.
The system also enables users to interrogate the last 12 months of marketing submissions, using the summaries to support pitches for new work or awards entries, in a process that Kell says makes it easier to provide consistent information about core business services and the practice’s approach.
ChatGPT has gained notoriety for sometimes presenting false information as fact. The information held in BDP’s version is curated and the practice controls what goes into the model, but Kell says users should not assume unchecked responses are 100% accurate. ‘It will take you 70 to 80% of where you need to get to, but people need to do the last 20% of the work,’ he said.
Plans to develop the tool’s capabilities include adding a dedicated button in Microsoft Teams to provide a simple point of access to the tool and potentially develop processes around responses to RFPs.
According to Kell, smaller practices lacking the R&D investment clout of BDP needn’t be put off using AI, as anyone with Microsoft Azure and software development skills could develop a similar solution, he said. He added that development for its version was not ‘hugely expensive’ compared to some of its licensing for second and third tier applications.