A law practice that became a laboratory.
Proprietary Legal Intelligence LLC grew out of one litigator’s decision to build serious AI inside his own firm instead of renting it from somebody else’s cloud.
Nick Mermiges, Esq.
Nick Mermiges is a South Carolina family-court litigator with more than eleven years in practice — custody, divorce, support, and the document-heavy financial disputes that come with them. Family law is among the most confidentiality-sensitive corners of the profession: the files hold children’s records, medical histories, and the complete finances of both sides of a marriage.
That sensitivity is why, when he decided his practice needed AI, he refused to send client files to outside services — and built the alternative instead: a production local-AI stack that his firm operates daily. A multi-GPU inference fleet on premises. A document-automation system that turns drafts into deterministic, court-ready filings. Retrieval across decades of his own work product, so the machine drafts in his voice. And pruning and evaluation pipelines that decide, with evidence, which models deserve a place on the rack.
In 2026 he registered Proprietary Legal Intelligence LLC to do two things with that experience: build the same class of system for other firms, and publish open-weights models and tools so the profession can verify the work.
From dictating letters to running a GPU fleet.
-
The awakening
A response that should have taken five hours
An AI model produced, in moments, work that would have cost a long afternoon — impressive and unsettling at once. The choice was to surf the wave rather than be swept under it, but on strict terms: use AI only where its work could be reviewed and verified.
-
The constraint
The models come to the files
Sending client files to a cloud API was never an option for family-court matters. So the architecture inverted: the models run in the office, and the client data never leaves it.
-
The build
A harness, not a chatbot
Recurring work decomposed into narrow, checkable steps; deterministic document builds; retrieval over a decade of the firm’s own filings; and fact-check gates that fail loudly instead of hallucinating quietly.
-
The company
From a hardware post to a two-hour talk to a business
A post about the home-office server drew engineers from a major legal research company, who asked for a two-hour technical talk. Their questions made the case that lawyers and engineers were each missing half the picture — and that became Proprietary Legal Intelligence.
The firm in brief
- Company
- Proprietary Legal Intelligence LLC (“PLI”), founded 2026
- Founder
- Nick Mermiges, Esq. — South Carolina family-court litigator, 11+ years in practice
- Two arms
- PLI Consulting (private AI for firms) and PLI Labs (open-weights model prunes and open-source tools)
- Where
- 1720 Main Street, Suite 202, Columbia, SC 29201
- Online
- ProprietaryLegal.ai · ProprietaryLegal.com · ProprietaryLegalIntel.com · GitHub · Hugging Face
What we hold to.
Confidentiality by architecture
Privacy promises expire; architecture doesn’t. If client data physically cannot leave the building, no policy change, breach notice, or acquisition on a vendor’s side can put it at risk.
The lawyer stays in the chair
Machines assemble, retrieve, and draft. Lawyers decide. Every workflow we design ends in attorney review, and we will not build systems meant to skip it.
Fail loudly
A legal AI system that guesses silently is worse than none. Our pipelines are built to refuse and flag when they cannot verify, rather than produce confident nonsense for a court file.
No lock-in, ever
You own the hardware, the models are open weights, and the documentation is yours. A PLI engagement should end with your firm more independent, not less.