What if the best AI chatbot is not a chatbot at all, but a constrained search engine?

Prerequisite

Basic understanding of:
Drupal content architecture
Search concepts (Solr, Elasticsearch, or similar)
General awareness of LLMs

Target Audience

Drupal site builders, webmasters, content managers, and developers working with Drupal as a CMS for public or private sector websites. It is especially relevant for professionals interested in understanding how AI can support content management, editorial workflows, accessibility, and user experience within Drupal-based platforms. Attendees looking for practical ideas to integrate AI into existing digital ecosystems will benefit most.

Outline

EPSO’s AI assistant EPSY is not a chatbot in the usual sense. It is a constrained AI search interface designed for public sector use.

In this session, we explain EPSY, the assistant of the European Personnel Selection Office, and show how it solves a hard problem: guiding tens of thousands of candidates through complex, multilingual recruitment flows without breaking trust or compliance.

We analyze its design choices: prompting, strict query framing, and controlled outputs.
Key topics include AI search vs chat, designing for accuracy over fluency, and how to operate AI in regulated environments.

You will leave with a clear mental model and practical patterns to implement AI-assisted search in Drupal projects.

Learning Objectives

Learning Objectives
Understand the difference between chat UX and AI search UX
Learn how EPSY reduces hallucination risk through design
Identify when to use stateless vs conversational AI
Map RAG patterns to Drupal architectures
Apply guardrails for AI in regulated environments

Experience level
Intermediate