AI agents accelerate legal dossier preparation for government recovery operation
Streamlining the preparation of complex legal objection dossiers through multi-agent orchestration and rigorous human oversight.
15 min, no strings · No sales pressure
Bezwaarschriftenadviescommissie (BAC)
The Bezwaarschriftenadviescommissie (BAC) handles complex objections within the Dutch childcare-benefits recovery operation. SevenLab was tasked with building an AI agent platform to assist legal caseworkers in navigating massive, multi-year dossiers to prepare structured assessments. The goal was to increase the speed and accuracy of the preparation phase while maintaining strict compliance with the EU AI Act.
Caseworkers faced the daunting task of manually reviewing up to 40 documents per case, spanning 14 years of financial and family history. This manual extraction of timelines and claimant accounts was slow and prone to inconsistency across thousands of cases. Furthermore, creating a legally sound assessment required cross-referencing dozens of standard considerations, making the process difficult to audit and scale without technological intervention.
A look inside
What SevenLab built
We implemented a sophisticated agent pipeline using Mastra orchestration and EU-hosted LLMs to automate the heavy lifting of document analysis and drafting. The system uses chapter-aware extraction agents to build structured facts and generation agents to draft factual summaries grounded in a legal knowledge base via RAG. Every AI-generated line includes direct citations to source documents, followed by an automated AI fact-check pass to ensure data integrity before mandatory human review.
Chapter-aware fact extraction
Automatically pulls structured timelines and family data from multi-format dossiers using specialised extraction agents.
RAG-grounded legal drafting
Generates assessments based on previously accorded cases and a legal knowledge base to ensure consistency.
Automated AI fact-checking
A secondary agent audits drafts against original sources to prevent hallucinations or errors in dates and amounts.
Audit-ready citation engine
Every drafted sentence includes clickable links to the specific source document and page for full transparency.
Measurable business impact
The platform has significantly reduced the time required to produce a first-draft factual summary and assessment, allowing the committee to focus on final decision-making. By providing per-case confidence scoring and full auditability, the system ensures that high volumes of objections are handled with unprecedented consistency and legal rigour.
The platform has transformed how we handle complex dossiers by providing a high-quality first draft in a fraction of the time. The mandatory human-in-the-loop design ensures that our committee remains the sole decision-maker, supported by transparent, auditable AI insights.
Share
Related projects
Ensuring maritime safety with an AI assistant for gas tankers
Revolutionising Dutch absence management with an integrated AI platform
Digitising safety training for a multilingual industrial workforce
Trusted by teams at
Want results
like these?
Tell us your challenge and we'll show you how we'd solve it, with a clear scope, timeline, and fixed price.
Talk directly with our AI specialists