Digital Employee · Lawyer
Justine.
The lawyer. Powers JustineAI, the Agentic Legal Operating System — six editions on one Eve-Legal F5/reasoner. Personal Injury in market today; five more on the way.
Not a legal-research tool — an AI with the agency of a lawyer: the reasoning to work a case, the memory to recall every fact and filing across its lifecycle, and the reach to carry the intake, practice-management, and portfolio-scale work a lawyer carries, not just the legal analysis.
Justine is the Digital Employee for legal services. She is the runtime expression of the Eve-Legal compound reasoning model — an F5/reasoner-generation Eve-Fusion architecture trained on the Eve-Genesis Legal Edition synthetic reasoning corpus. Five cooperating models compose per request: Microsoft Phi-3 as classifier, the Phi-4-derived Eve-Legal small reasoning model as the workhorse, three frontier models as the frontier slots, one of them a long-context model providing the 10-million-token longitudinal context that lets Justine read an entire case file in a single thought.
Justine powers JustineAI, MindHYVE's Agentic Legal Operating System. JustineAI ships in six editions on one shared reasoner. The Personal Injury edition (PI) is in market today — 31 capabilities across the nine-status case lifecycle, with Justine's Workspace v1.2 for intake (per-field provenance and validator council), the nine-step Curation Layer over up to 50,000 pages of records, the four-pass demand-letter pipeline with CourtListener citation verification, and the one-click Demand Package assembly. Mass Tort (MT) and Workers' Comp (WC) are next; Medical Malpractice (MM), Insurance Bad Faith (IB), and Employment Class (EL) follow on the roadmap. The architectural moat is the supervisor pattern: one Justine coordinates stage-specialized sub-agents — intake, medical, valuation, strategy — across the case, and at MT scale, thousands of plaintiff sub-agents in a single reasoning context.
Justine is attorney-attested by design. Every output is work product, the must-review-before-final gate is non-dismissible, and attorney-client privilege is preserved by structuring every operation under the supervision of the attorney of record. Audit logs record every reasoning step, decision, and revision — retained for the contractual period and exportable for litigation discovery, malpractice insurance, and ethics review. She inherits the universal Digital Employee properties: continuous learning through deployment, infinite memory recall, compliant posture engineered into the substrate, deployment on Eve-Grid on Microsoft Azure, US data residency by default, ISO 27001 / SOC 1/2/3 / HITRUST inherited from the Azure foundation, and 24/7/365 availability.
Not just the legal analysis. The whole practice.
A lawyer is never only a litigator. On a plaintiff matter alone, a lawyer carries several jobs at once — intake counselor, case curator, damages valuer, strategist, and practice operator. Justine's agency spans all of them, in three orders of capability. First-order casework is where it starts, not where it stops.
First order
Casework
The legal work on the case itself: intake and fact curation through a nine-step Curation Layer over up to 50,000 pages of records, medical-chronology synthesis, damages and settlement modeling, four-pass demand-letter drafting with CourtListener citation verification, and one-click Demand Package assembly.
Second order
Practice management & operations
The work around the case: deadline and calendar tracking, conflict-of-interest checking, lien negotiation and trust accounting, time entry and billing, the client portal, and file assembly for e-filing and e-signature — integrated with Microsoft 365 and Filevine.
Third order
Portfolio-scale orchestration
The supervisor pattern: one Justine coordinates stage-specialized sub-agents — intake, medical, valuation, strategy — across a case, and at mass-tort scale, thousands of plaintiff sub-agents in a single reasoning context, with bellwether scoring and settlement-matrix modeling across the plaintiff pool. The architectural moat for inventory work.
An AI with agency — and the architecture to wield it.
Justine's agency is broad by design. What makes it deployable in regulated legal practice is the architecture beneath it and the Operating System above it.
Compound reasoning
Backed by Eve-Fusion™ F5
Every request runs through five models — a Phi-3 classifier, the Phi-4-derived Eve-Legal reasoner, and three composed frontier models, one a long-context model with a 10-million-token window that reads an entire case file in a single thought. Legal reasoning depth no single model delivers.
Eve-Legal F5 →Dedicated memory
Its own Azure AI Search
A dedicated Azure AI Search instance gives Justine infinite memory recall across a case's full record — intake through settlement. She carries every fact, filing, and deadline forward instead of forgetting it between turns.
Bounded by the OS
Agency in the employee, control in the OS
Justine reasons with full agency; JustineAI — the Operating System — sets the bounds: every output is attorney-attested work product, the must-review-before-final gate is non-dismissible, attorney-client privilege is preserved under the supervision of the attorney of record, and a tamper-evident audit log records every reasoning step. The power is the employee's; the guardrails are the OS's.
One request, five models, one answer.
Eve-Legal F5 is a compound, not a single model. Press Run to watch a request move through it — or click any stage to inspect it.
Stage 1 / 8 · Request
A user asks
A request enters through the Digital Employee’s Operating System — the surface that frames, scopes, and governs everything that follows.
The AI reasons. The human decides. The OS records.
“Agency in the employee, control in the OS” is the most important claim on this page — and the most abstract. Step through it.
Step 1 · Agency
Justine reasons
Justine works the case with full agency — reasons through it and drafts the demand letter. No human in the loop yet.
PI in market today. Five more editions, one Operating System.
Six editions on one Eve-Legal F5/reasoner. The reasoning core, the supervisor-over-sub-agents pattern, and the case-management OS layer (deadlines, conflict-check, trust accounting, time tracking, e-filing, e-signature, client portal, Microsoft 365 and Filevine integration) are shared across every edition. What changes per edition is the calibration — workflow, document pipeline, jurisdictional rules, regulatory taxonomy.
JustineAI PI
Plaintiff-side personal-injury practice
Personal Injury edition. Justine's Workspace v1.2 with per-field provenance and validator council; nine-step Curation Layer over up to 50,000 pages of medical records; four-pass demand-letter pipeline with ICD-10/CPT validation and CourtListener citation verification; one-click Demand Package assembly (cover letter + Medical Chronology + Bill Stack). 31 capabilities across the nine-status case lifecycle. 50 states + DC; top-10 PI states with curated case-law packs.
JustineAI MT
Mass tort and MDL specialists
Mass Tort edition. One Justine supervisor coordinates thousands of plaintiff sub-agents in a single reasoning context — bellwether scoring, settlement-matrix modeling at scale, lien negotiation across the plaintiff pool, plaintiff fact-sheet auto-population. The supervisor + sub-agent pattern is the architectural moat for inventory work.
JustineAI WC
Workers' compensation (applicant and defense)
Workers' Compensation edition. State-specific filings (CA DWC 5021/PR-2, NY C-4, FL DWC-25, TX DWC-69), AME/PQME analysis, MMI tracking, MTUS/ACOEM treatment-utilization review, permanent-impairment rating modeling, Medicare Set-Aside coordination. ChironAI Occupational Medicine is the upstream feed — cross-portfolio reasoning, not a hand-off.
JustineAI MM
Medical malpractice litigation
Medical Malpractice edition. Multi-provider chronology across 5,000–50,000 pages per case in a single 10M-token context — the long-context model reads the entire record set in one thought. Causation reasoning anchored to cited standard-of-care deviations, Daubert challenge preparation, certificate-of-merit drafting, expert-witness management.
JustineAI IB
Plaintiff-side insurance bad-faith litigation
Insurance Bad Faith edition. Carrier policy-language interpretation clause-by-clause, claim-file analysis assembled into a jury-ready narrative, claim-handler deposition preparation against prior testimony and the file's own contradictions, Brandt fees calculation. Prior-claim pattern detection across the carrier's portfolio when public.
JustineAI EL
Plaintiff-side employment class action and FLSA
Employment Class edition. Rule 23 / FLSA collective certification motion practice, damages modeling at class scale from production data, FLSA fluctuating-workweek calculations layered with liquidated damages and prejudgment interest, opt-in conversion tracking, EEOC charge orchestration at class scale.