The problem
A legal case management platform handling workers' compensation workflows (WCAB, QME, CNR) needed attorneys to manually re-read scanned filings and case histories before every decision — slow, repetitive, and easy to get wrong on complex cases.
What we built
Built an agent on the Claude API that doesn't just answer questions but takes action through tool calls, connected to case data, document stores, and internal tools via the Model Context Protocol (MCP). A RAG pipeline grounds every answer in the actual case record instead of model memory, fed by an OCR pipeline (Tesseract) that turns scanned legal filings into structured, searchable text. The agent runs on a microservices backend with audit logging and cron-based automation, so every action it takes on a case is traceable.
The result
Reduced the manual review load on attorneys handling high case volumes, with every agent action logged and auditable — built as one of three core platforms owned end-to-end by the team.
Delivered by NovaBuildz engineers before the studio was founded — the same team now works under one roof.
Start a similar project