The Challenge
Wolters Kluwer needed to integrate Generative AI into their Schulinck knowledge base, the gold standard for Dutch municipalities, without compromising on reliability. Legal questions regarding social domains (like the Participatiewet or WMO) are complex and dynamic; an incorrect answer can have serious consequences. The challenge was to build a system that could "reason" over vast legal texts and provide legally sound answers with citations, rather than just plausible-sounding summaries.
Our Approach
We moved away from standard "out-of-the-box" AI implementations and established a rigorous, quality-first and experimentation driven engineering culture:
- Deep Tech & RAG Architecture: We built a custom Retrieval-Augmented Generation (RAG) infrastructure that allows for easy switching and rigorous testing of different Large Language Models (LLMs). This insulated the product from vendor hype and ensured we only deployed models that actually performed better on legal reasoning.
- Automated Evaluation Framework: Instead of relying on subjective feedback ("it looks good"), we built an automated evaluation framework to measure answer quality objectively against a gold standard.
- Human-in-the-Loop Development: We integrated Schulinck’s legal experts directly into the product team. This created a daily feedback loop where engineers and lawyers worked side-by-side to fine-tune prompts and context, treating the AI as a "CoPilot" that knows when to say "I don't know."
The Results
The Schulinck AI-Desk is now a live, trusted tool used by municipalities across the Netherlands.
- High User Trust: By prioritizing accuracy over raw speed, the tool achieved overwhelmingly positive feedback from legal professionals who rely on it for complex casework.
- Verifiable Accuracy: Every answer includes direct citations to the source material (legislation, jurisprudence, policy), allowing users to instantly verify the AI’s output.
- Workflow Transformation: The tool doesn't just search; it acts as a reasoning partner, helping municipal workers draft better decisions faster and reducing the time spent digging through complex regulations.
Deep Dive
For a candid look at the development process, read our joint retrospective from the Data Expo where we share the hard lessons learned: Read: 10 Pitfalls of Developing Impactful AI (and How to Prevent Them)
