Proprietary Legal Intelligence

Proprietary Legal Intelligence

Your firm already built the intelligence.

Decades of briefs, matter files, research, and hard-won judgment sit scattered across drives, inboxes, and memories. We build the systems that connect it to frontier AI, so the work comes back in your organization’s context — and under your attorneys’ review.

See it come together Start a conversation

Scroll to watch scattered work product become a working system. Every matter shown is fictional.

  1. 01 Scattered

    It’s all already here.

    Pleadings, discovery, transcripts, research, forms, correspondence — and what experienced attorneys simply know. It’s organized by where it happens to be stored, not by what it means. So every AI conversation starts from zero.

  2. 02 Structured

    Documents resolve into matters.

    Each file is placed in its matter, dated, typed, and traced to its source. The archive stops being storage and becomes a record.

  3. 03 Connected

    Matters reveal issues. Issues connect to evidence and authority.

    The same questions recur across years of work. Connecting them turns one matter’s research into the next matter’s head start.

  4. 04 Standards

    Your best work becomes the standard.

    The briefs and agreements your firm is proudest of become exemplars. Attorney corrections become standing preferences. Steps done the same way every time become workflows.

  5. 05 Context

    A request arrives. The right context assembles.

    “Draft our opposition to the motion to compel in Harper.”

    The system gathers what a senior colleague would pull: the matter’s evidence, the controlling authority, the firm’s strongest prior oppositions, and this attorney’s preferences. Nothing else.

  6. 06 Checked

    The model drafts. The record checks it.

    A frontier model reasons over the assembled context, one workflow stage at a time. Each factual statement is traced to a source. Anything unsupported is flagged, not quietly shipped.

  7. 07 Reviewed

    An attorney decides. The firm learns.

    A lawyer reviews, edits, and signs off. The edit doesn’t disappear into a chat history — it returns to the knowledge base as a preference, so the next draft starts closer.

The difference

The model is one part of the system.

Frontier models reason and write extraordinarily well. What they don’t have is your matter, your record, your precedent, or your standards. Supplying that context — and checking the result — is the work.

A chat window

Generic prompt. Generic model. Generic answer.

Draft excerpt

Plaintiff respectfully opposes Defendant’s Motion to Compel. Courts routinely deny such motions where the moving party has failed to meet and confer in good faith. See Hollis v. Brandt, 1042 F.3d 118 (4th Cir. 2019). Plaintiff has produced all responsive documents in its possession, custody, or control…

  • ×Citation checked against nothing. (F.3d never reached volume 1042.)
  • × Asserts facts it was never given.
  • × Written in no one’s voice.

Knows What you typed.

A system built on your work

Your request, assembled with everything it depends on.

Draft excerpt

Defendant served its Second Requests for Production on March 3. Plaintiff served written responses within thirty daysEx. 4, and counsel conferred by letter on April 11Ex. 6, offering a rolling production schedule that Defendant accepted in writingEx. 7, p. 2. Defendant has raised no deficiency since.

  • ✓ Every fact pinned to the page it came from.
  • ✓ Structure drawn from the firm’s strongest prior oppositions.
  • ! Unsupported statement held for attorney review.

Knows The matter, the record, your precedent, your standards.

Illustrative excerpts from a fictional matter. The citation at left is deliberately invented to show how an unchecked answer can fail.

Already using AI?

Access was the easy part.

Most legal organizations now have ChatGPT, Claude, or Copilot. The harder problem is making AI reliably understand the organization — without pasting documents into a chat window and re-explaining the matter every time.

Knows The organization — and learns from review.

Knowledge, context, process, validation, and attorney feedback compounding in one system. Every matter makes the next one better.

What each stage of legal AI can draw on
Capability0102030405
Your prompt
Documents you paste in
The firm’s archive, searchable
The matter record, with sources
Exemplars and standards
A defined, staged process
Checks against the record
Learning from attorney review

Each stage keeps what came before it. Most of the value is on the right side of this table — so is most of the work.

Who it’s for

One idea. Five kinds of legal work.

The thesis doesn’t change: build AI around the organization’s own knowledge. What changes is where that knowledge lives and what the work demands.

I Civil litigation, trials & appeals

From the case file to the argument — at trial and on appeal.

A large case file becomes a chronology. The chronology surfaces the issues. The issues pull their evidence, expert opinions, and authority, and the argument is built from all of it — for the dispositive motion, the trial notebook, or the appellate brief, with every assertion pinned to the page of the record it came from.

  • Chronologies & evidence extraction
  • Expert discovery: reports, depositions, challenges
  • Deposition & transcript summaries
  • Discovery review & deficiencies
  • Motions & trial preparation
  • Record on appeal & appellate briefs

II Law firms & practice groups

Make the firm’s memory available to every matter.

Years of matters — and the judgment of every partner, associate, and paralegal who worked them — spread across a document system, shared drives, inboxes, and people. We connect it so precedent, research, and prior work surface where the next matter needs them, and stay with the firm when people move on.

  • Precedent retrieval
  • Research reuse
  • Matter summaries
  • Practice-group standards
  • Training & delegation
  • Quality control

III Individual attorneys & small firms

Stop re-explaining your practice.

Your matters, your jurisdiction, your arguments, your structure, your voice. Instead of briefing a chatbot from scratch, the system assembles that context for the task — and returns a draft that reads like your work because it’s built from your work.

  • Drafting in your own voice
  • Client correspondence
  • Pleadings & forms
  • Document generation
  • Meeting notes to case memory
  • Intake to first draft

IV Corporate legal departments

Answers grounded in approved positions.

Contracts, policies, correspondence, regulatory material, and outside-counsel work product, organized so routine requests are answered from the department’s approved positions rather than a model’s general knowledge — and exceptions reach counsel with the reasoning attached.

  • Contract review against playbooks
  • Policy analysis
  • Internal legal requests
  • Investigations
  • Document comparison
  • Outside-counsel knowledge

V High-volume & repeatable work

Not one giant prompt.

When matters differ but the analysis repeats, the work can be decomposed into stages — and each stage gets the right tool: a model, a retrieval step, deterministic code, a validator, or a person.

  • Intake & triage
  • Record & financial analysis
  • Demand & response letters
  • Recurring filings
  • Review at scale
  • Quality control

Also Legal-technology teams and model builders: we collaborate on legal evaluation, workflow design, and open legal models. See PLI Labs →

The architecture

The model is the foundation, not the building.

  1. FoundationModels & compute. Frontier models through APIs your organization approves, or open-weight models on hardware you control — chosen per task, replaceable as they improve.
  2. StepsYour record. Documents, matters, and data, indexed with provenance and the access permissions they already carry.
  3. ColumnsThe system. Knowledge, context assembly, exemplars, staged workflows, validation, and memory of attorney corrections.
  4. EntablatureAttorney review. Professional judgment spans the whole structure. Nothing consequential leaves without it.
  5. PedimentWork product an attorney can verify, sign, and stand behind.
A courthouse drawn in circuitryCircuit traces rise from a processor at the foundation, pass through the steps, and become the fluted Corinthian columns of a courthouse. The six columns are labeled knowledge, context, exemplars, workflows, validation and memory. The entablature is inscribed work product, precedent, judgment, and the pediment holds a small graph of connected knowledge.WORK PRODUCT · PRECEDENT · JUDGMENTKNOWLEDGECONTEXTEXEMPLARSWORKFLOWSVALIDATIONMEMORY

Deployment & control

Run it where the work requires.

Sensitivity varies by matter and by client. The architecture is the same whether a workload uses a frontier model through an API your organization has vetted, a private cloud deployment, or open-weight models on hardware you own.

  • Permission-aware retrieval

    Retrieval can be scoped to what the requesting person is already allowed to see.

  • Provenance

    Outputs trace to the documents and pages they relied on.

  • Run records

    Each workflow run can record its inputs, stages, checks, and reviewer.

  • Human review

    Required before anything consequential leaves the system.

  • No lock-in

    Portable across model vendors. The workflows and documentation are yours.

What we won’t claim No architecture resolves your confidentiality, privilege, or regulatory obligations for you. Those judgments stay with your lawyers. We make the tradeoffs explicit and design around the obligations you already carry.

Who builds it

Built by a trial lawyer. Run on real cases.

Proprietary Legal Intelligence was founded by Nick Mermiges, a South Carolina trial lawyer with more than eleven years in practice — civil litigation, civil trials and appeals, complex and expert discovery, and family law. A University of Michigan graduate, he finished in the top 5% of his class at the University of Miami School of Law, where he served on the Law Review and was elected to the Order of the Coif.

The systems described here began in his own practice — drafting, discovery analysis, financial tracing, transcript summaries — and run on live matters under real deadlines, with a lawyer reviewing every output.

A post about that setup led engineers at a major legal research company to request a two-hour technical talk. Their questions made the case for the company: lawyers and engineers were each missing half the picture.

“Your proprietary work product is your proprietary intelligence.”Nick Mermiges, Esq. — Founder
About the founder Open research
Nick Mermiges, Esq., founder of Proprietary Legal Intelligence, in a navy suit and red tie in front of shelves of legal reporters
Nick Mermiges, Esq.Founder · Columbia, South Carolina

How it starts

Two and a half days. A clear answer.

We review your forms, processes, software, and files under an NDA. We train your attorneys to use AI far more deeply. Then we lay out what building further would take, with a fixed fee for each option.

  1. 01 · Review

    How your firm works

    Forms, templates, processes, software, and a sample of files, under a mutual NDA. You decide what we see.

  2. 02 · Train

    Your attorneys, hands on

    Deeper, verifiable use of the AI tools you already have, on your own kind of work. Valuable on its own.

  3. 03 · Options

    What to build next

    Findings and concrete options for firm-wide deepening, building, and automation, each with a fixed fee.

  4. 04 · Build

    Only if it’s worth it

    Built and tested on past matters before it touches live work. No lock-in; the roadmap is yours either way.

Working session · video$1,00090 minutes, prepaid. Credited toward on-site.
On-site assessment & training$15,0002½ days, 20 hours. Travel at cost.
Build & automationFixed feeQuoted from the assessment.

Schedule of fees and what’s included

The last mile is yours

The system prepares. The lawyer argues.

AI can assemble the record, draft the brief, and check the draft against its sources. Standing up in court is still your job. We want you to walk in better prepared.

nick@proprietarylegal.com

Two paragraphs about your practice is enough for an honest first read. Please describe the shape of the work, not client confidences.