Jurimetria AI

Law firms and legal departments need to track cases and understand historical outcomes by court, district, judge and type of claim — information that lives inside the text of the rulings.

Case, party and claim management; ruling text sent for AI analysis with structured output; and a jurimetrics dashboard computed by the database. Core principle: the AI reads one document at a time and never computes statistics.

• React + Vite frontend → Node.js/Express API → PostgreSQL. The AI key exists only on the backend.
• Layered backend (routes → controllers → services → repositories) plus an isolated AI layer, which made it possible to switch providers without touching controllers or services.
• Asynchronous analysis: the record is created as Pending before the external call, the API answers 202 and the frontend polls the status.
• Structured Outputs with a Zod schema, validated again on the backend before persisting.

• CRUD for cases, parties, claims, rulings and appellate decisions, plus supporting catalogs (courts, districts, judges).
• PDF and DOCX ruling upload with text extraction.
• Anonymization (LGPD) of CPF, CNPJ, email, phone, postal code, ID and banking data before anything is sent to the AI.
• Jurimetrics dashboard by judge, by claim and by court.

• No ruling is left without a recorded analysis: if the AI call fails, the status becomes Error with a log and the user can retry.
• Separating semantic reading (AI) from aggregation (SQL), so the numbers are always descriptive, auditable statistics.
• Automated tests with node:test that never call the real API or require a running database.