How we implemented AI search on GovCMS
Track: Technical
Keyword search matches words, not meaning. On Cancer Australia's site, packed with clinical guidance, statistics and publications, that can be a problem. People search by topic, not by whatever exact term the source document happens to use.
We built semantic search into GovCMS. Utilising hybrid queries that blend vector and keyword scoring, with an on-demand AI summary sitting on top, backed by OpenSearch and AWS Bedrock.
This talk covers the architecture: Drupal's Search API in front, with two separate search pipelines so Retrieval-Augmented Generation (RAG) latency never blocks the initial results. It also covers the security work: moving OpenSearch off of Lagoon and onto Amazon's managed offering, then locking it behind CloudFront, WAF and a private subnet.
We'll walk through the build itself: the contrib modules we used, the infrastructure calls we made, and what we'd tell any agency doing this.
Keep the AI answer assistive, not authoritative. And don't let new infrastructure become the exception to your security model.
We built semantic search into GovCMS. Utilising hybrid queries that blend vector and keyword scoring, with an on-demand AI summary sitting on top, backed by OpenSearch and AWS Bedrock.
This talk covers the architecture: Drupal's Search API in front, with two separate search pipelines so Retrieval-Augmented Generation (RAG) latency never blocks the initial results. It also covers the security work: moving OpenSearch off of Lagoon and onto Amazon's managed offering, then locking it behind CloudFront, WAF and a private subnet.
We'll walk through the build itself: the contrib modules we used, the infrastructure calls we made, and what we'd tell any agency doing this.
Keep the AI answer assistive, not authoritative. And don't let new infrastructure become the exception to your security model.
Speakers
Adam Bramley
Adam has been an active member of the Drupal community since he began building websites in 2010, accumulating experience across technical leadership, architecture, engineering and consulting for government and large enterprise clients. He has built a strong reputation for delivering consistently high quality and maintainable websites.
At PreviousNext, Adam works as a Lead Developer, bringing expertise across a broad range of Drupal integrations, testing frameworks and open source methodologies. For the Drupal open source project, Adam is a subsystem maintainer for Block Content and Node. His meticulous approach to code quality ensures the solutions he delivers are built to last.
At PreviousNext, Adam works as a Lead Developer, bringing expertise across a broad range of Drupal integrations, testing frameworks and open source methodologies. For the Drupal open source project, Adam is a subsystem maintainer for Block Content and Node. His meticulous approach to code quality ensures the solutions he delivers are built to last.