NJ BriefBank uses domain-specific AI to aid overburdened public defenders

Researchers from Princeton and other institutions, in partnership with the New Jersey Office of the Public Defender, have developed NJ BriefBank—a retrieval tool designed to help overworked public defenders quickly find relevant appellate briefs. The system tackles a critical gap: while AI tools are often proposed for public agencies, little evidence existed for how they could meaningfully support day-to-day defense work. The team found that existing legal retrieval benchmarks fail when applied to real public defense queries, so they injected domain knowledge—including query expansion with legal reasoning, domain-specific data, and curated synthetic examples—to dramatically improve retrieval quality.

To facilitate further research, the authors released a taxonomy of realistic defender search queries and a manually annotated evaluation dataset that closely correlates with proprietary annotations from experienced public defenders. This work highlights a pragmatic, human-centered approach to applying AI in a high-stakes public interest setting, moving beyond generic benchmarks to address the unique constraints of under-resourced legal offices. The result is a tool that could reduce research time and help defenders provide better representation to clients facing overwhelming caseloads.