A legal reasoning engine that reads intake documents like a paralegal

The problem

Law firms drown in intake. Every new matter arrives as a stack of documents (often scanned PDFs from the client's own files), and a partner or paralegal reads all of it to extract the case-relevant facts: parties, dates, amounts, prior counsel, jurisdictional triggers. Hours per matter, every matter.

The deeper problem isn't time. It's that the partner reading is the most expensive person in the firm. Hourly rate going to a task an AI can do in seconds, instead of the work that requires legal judgment.

What I did

Three-month engagement, multiple firms onboarded.

• End-to-end automation across intake, case management, and analytics. Not a point solution. Pipeline starts at first contact, ends with a complete structured matter file in case management.
• n8n + Claude 3.5 Sonnet for document extraction. Returns structured JSON: parties, dates, amounts, jurisdiction, summary, prior counsel, flags. n8n routes it into the firm's CRM + case management.
• Custom dashboards. Intake throughput, conversion to engaged matters, bottlenecks at a glance.
• Discovery with firm owners. Architecture pattern shared, configuration bespoke per firm.

The detail that mattered: lawyers don't trust AI summaries by default, and they're right not to. The system surfaces the structured extraction AND the raw passage that produced it, side by side. Partner verifies in seconds. Trust built by transparency, not by accuracy claims.

Outcome

• Hours per week back into casework instead of intake.
• Matters move through the early pipeline faster.
• Partners stop being human OCR. They review extractions, not documents.
• Dashboards give firm leadership real operational visibility for the first time.