Content editors at the European Commission manage hundreds of websites in 24 languages. Every publication needs to be accurate, readable, and compliant with policy, and it needs to stay that way as policies change. Doing that manually across thousands of pages does not scale. During the EC's Drupal AI Hackathon, our team built an AI content validation engine that helps editors run structured checks, review AI suggestions in an isolated environment, and approve changes on their terms. No auto-changes, no data leaving the stack. This session walks through the problem, a demo, and the architectural decisions behind a solution where AI assists and humans stay in control.

Prerequisite

- General knowledge of Drupal site building and content editing
- Basic concepts of AI and Agentic AI
- Familiarity with concepts like entities, forms, or APIs
- No prior experience with AI required

Target Audience

Agency leaders, solution architects, innovation teams, technical strategists

Outline

At the European Commission, content editors manage hundreds of websites in 24 official languages. The real challenge is not creating content but validating it: checking facts against policy, enforcing SEO standards, maintaining accessibility, and ensuring consistency.

During the EC's Drupal AI Hackathon in January 2026, our team built an AI-powered content validation engine that helps editors validate, improve, and monitor content across its lifecycle. The solution won first place.
This session covers the problem, the hackathon challenge, a demo, and a technical walkthrough of the architecture behind it.

You will see how we built an AI assistant that evaluates content against configurable editorial rules covering fact accuracy, readability, SEO, and policy alignment. The assistant works in an isolated environment where editors review suggestions before anything touches published content. Every change requires explicit human approval, enforced deterministically outside the AI layer.

On the architecture side, we explain why we chose to let Drupal orchestrate the AI steps rather than delegating that to an LLM agent and how that decision shaped the entire solution. We also cover async execution, vendor-independent model integration, and data sovereignty by design.
If you are building agentic AI workflows in Drupal, this session gives you a practical inspiration: what worked, what didn't, and the principles.

Learning Objectives

- How to architect agentic AI workflows in Drupal where the CMS orchestrates the process and the LLM executes discrete steps, and why this matters for cost, speed, and reliability
- A practical approach to human-in-the-loop design that is enforced deterministically, not by prompting an AI to "always ask for approval"
- How to build AI content validation that combines rule-based checks with contextual AI evaluation, using each where it performs best.
- Strategies for keeping agentic workflows data sovereign.

Experience level
Intermediate