Synopsis
CPRD.com combines Drupal Search API, Azure AI Search and a Foundry IQ knowledge base so people can ask questions about published content and receive answers with source references. This session follows a question from the Drupal interface through retrieval to a cited answer, and shares the design decisions behind content selection, answer quality and handling questions the content cannot answer.
Description
A search results page helps people find content. Sometimes they need an answer drawn from that content - for example, which approved studies discuss a particular health outcome, or what CPRD’s guidance says about applying for data access. They also need to see where the answer came from.
CPRD.com builds on its Drupal Search API and Azure AI Search integration with a Foundry IQ knowledge base. Published Drupal content is indexed for retrieval; the knowledge base uses that content to respond to natural-language questions; and Drupal presents the answer alongside references to its sources. Conventional search remains available when a results list is the better way to explore.
I’ll walk through this feature using representative research questions. The session will cover how content reaches the index, how retrieval and answer generation work together, how citations are returned and displayed, and how the experience handles insufficient evidence. I’ll also share the application and operational decisions behind the feature, including content boundaries, access, logging and evaluation.
Attendees will leave with a practical model for adding grounded question answering to a Drupal site that already has structured, searchable content.
Session outline
- The user journey: From a research question to an answer readers can verify.
- The Drupal search foundation: Search API, indexed content and Azure AI Search.
- How the knowledge base retrieves evidence: Follow a question through query planning, retrieval and answer generation.
- Answers and sources in Drupal: Citations, insufficient-evidence responses and the interface.
- Design and operational lessons: Content scope, access, evaluation, logging and cost.
- Questions.