Name in original language
NugepIA: Sistema de Gestão de Precedentes Judiciais Potencializado por IA
Initiative overview
NugepIA was developed by the Court of Justice of Rondônia (TJRO) to resolve critical bottlenecks in the management of binding precedents, such as IRDR and IAC, within the court’s NUGEPNAC unit. Prior to its introduction, monitoring of precedents relied on manual spreadsheets and institutional memory, creating fragmented communication, high risk of error and delays in reporting information to the National Council of Justice (CNJ). Judges’ chambers also lacked automatic visibility into suspended cases, negatively affecting productivity and legal certainty.
The initiative’s core objective is to automate the NUGEPNAC workflow using the Collectivus AI tool. The system automatically extracts key procedural dates—such as suspension, judgment and finality—directly from case documents, ensuring accurate and up‑to‑date records. It applies textual‑similarity analysis to group identical or highly similar cases, alerting judges in their workspaces to repetitive litigation and suggesting links to existing or potential binding precedents. In addition, the system performs daily monitoring of the Supreme Federal Court (STF) website to track updates in constitutional‑control actions relevant to ongoing cases.
Development progressed from initial mapping in February 2025 to full implementation by December 2025. Integration was completed across the PJe system, the Nugep System and the Chamber Module, eliminating manual reconciliation between platforms. This milestone institutionalised AI as a support tool for judicial decision‑making and laid the technological foundation for the broader AssessorIA platform, which now scales similarity‑analysis capabilities across the entire court.
Operationally, NugepIA eliminated manual data verification and enabled automated, accurate reporting to the CNJ’s National Precedent Bank (BNP). The AI achieves over 95% precision in grouping similar lawsuits and provides direct, actionable suggestions to judges’ chambers. Impact is measured through reduced screening and verification time, continuous validation cycles parallel to development and improved visibility of first‑ and second‑instance caseloads that were previously unmonitored by the central system.
Implementation addressed several challenges. Manual grouping of cases and spreadsheet‑based communication were error‑prone, and inconsistencies in procedural dates previously prevented automatic validation. These issues were resolved through full system integration and AI‑based date extraction, replacing manual review. Experience showed that successful AI use in the judiciary depends on coordinated planning between technical and legal teams, realistic timelines and continuous testing to mature similarity‑analysis functions.



























