JustineAI™ doesn’t run on a single foundation model. It runs on Eve-Legal F5/reasoner — a compositional fabric of cooperating reasoning models, deployed on Eve-Grid™ (Microsoft Azure). Plug-and-play. Provider-diverse. Domain-specialized. Designed so the constraints of one model never become the limits of your firm.
Four layers, one model. Numbered OSI-style — the foundational Substrate is Layer 01, ascending to the customer-facing Operating System at Layer 04. A request descends to the Substrate that grounds it; the answer ascends back.
The customer-facing product. The Personal Injury edition (and roadmap editions) run the institutional legal workflow a practice actually uses — structured intake, evidence, and demand construction.
The agent with agency — persistent identity, continuous learning, vector memory. Justine coordinates stage-specialized sub-agents. Justine reasons; the attorney decides.
Five cooperating models — a routing classifier, the domain-specialised legal reasoning model, and three frontier consultants composed best-fit per request — one of them for long-context analysis.
Eve-Grid™ on Microsoft Azure hosts the fabric; Eve-Genesis (Law Edition) trains the legal SRM via LoRA; Eve is the meta-orchestrator. The body, the cognition, the training.
Five cooperating reasoning models compose per request. The classifier routes; the legal SRM carries the domain workload; three frontier slots — deep synthesis, alternative reasoning paths, long-context analysis — add nuance and cross-validation, composed best-fit per request.
Sub-200ms routing across workload type, modality, reasoning intensity, and context-length needs.
Why this model occupies this slot — A lightweight classifier. The first responder of the stack — decides which downstream models compose for the matter at hand.
LoRA-fine-tuned on Eve-Genesis (Law Edition). The domain-specialized workhorse for the bulk of legal reasoning.
Why this model occupies this slot — Carries the majority of legal workloads at small-model cost and latency. Trained to reason like a senior litigator across canonical jurisdictional rules and procedural taxonomies.
Best-in-class long-form structured reasoning. Used when the matter is high-stakes and nuance matters.
Why this model occupies this slot — Frontier slot one. Strong at demand-letter narrative, complex damages reasoning, and matters requiring careful disambiguation of authority.
Frontier general reasoning. Complementary to the first frontier slot for cross-validation and ensemble routing.
Why this model occupies this slot — Frontier slot two. Provider diversity — when a client or jurisdiction prohibits a specific provider, this slot absorbs the constraint without rebuilding the agent.
Frontier slot three — the consultant for long-context analysis, composed best-fit per request like the other two.
Why this model occupies this slot — The third frontier consultant. It is a slot in the composition, not a promise about how much of a file is read at once: Justine’s document work is retrieve-and-cite — each record indexed passage by passage, each answer citing the passages it drew on.
The domain-specialised legal reasoning model is fine-tuned on Eve-Genesis (Law Edition) — a proprietary synthetic reasoning corpus calibrated to legal workflows. The dataset is 100% synthetic by construction. Your firm’s matter data is never used to train models.
Justine supervises stage-specialized sub-agents (intake, medical, valuation, strategy) across the case lifecycle. The sub-agents are NOT separately branded — Justine is the only Digital Employee with a name. This is the architectural moat for Mass Tort: one supervisor coordinating thousands of plaintiff sub-agents.
JustineAI™ runs on Eve-Grid™ — MindHYVE.ai™’s proprietary cloud architecture anchored to Microsoft Azure. The marketing site is on Azure Static Web Apps; the PI application is on Azure Container Apps, Functions, Cosmos DB, Storage, and Key Vault. Customer-managed keys available on enterprise plans.
Each firm’s matters live in an isolated Cosmos DB partition with per-firm encryption keys. The supervisor pattern enforces tenant boundaries at the orchestration layer — sub-agents cannot reach across firms.
The PI application runs in US Azure regions. Matter data stays in the customer’s region of record. International deployment is configurable but never automatic.
The Microsoft Azure platform layer holds ISO 27001, ISO 27018, SOC 1/2/3, PCI DSS, and HITRUST attestations. JustineAI™ inherits these at the platform layer; our own attestation roadmap is on the Trust page.
Firm users sign in with email and password (bcrypt-hashed); platform staff sign in with Microsoft Entra ID. Role-based, matter-scoped authorization on every action. Single sign-on for firms is on the roadmap, not available today. Per-user, per-matter, per-action authorization decisions logged.
TLS 1.2+ in transit. Azure-managed keys at rest; customer-managed keys are on the enterprise roadmap. No data is stored outside the customer’s Azure region of record.
Every reasoning step, every output, every user action logged to Azure Monitor. Logs are retained for the agreed contractual period and exportable for litigation discovery, malpractice insurance, and ethics review.
Service-to-service traffic stays inside the Azure backbone via private endpoints. No backend services have public ingress in production. Web Application Firewall policies and CSP headers enforced on every public surface.
The five-slot architecture in depth — classifier, legal SRM, three frontier slots — plus the supervisor pattern sitting one layer up.
The proprietary synthetic reasoning corpus that fine-tunes the legal reasoner. 100% synthetic by construction. The case for synthetic data in legal reasoning.
Tenant isolation, US data residency, and inherited platform attestations. The cloud that lawyers already trust.
Three real demand-letter excerpts; every citation a public case authority; every one verifiable on CourtListener in one click.
The compositional fabric visualised. Hover or click any slot to see the model, the role, and why that model occupies that slot.
Justine + six named sub-agents (Intake, Medical, Valuation, Strategy, Deposition, Case Advisor) — the tier above F5/reasoner.
Five anonymized demand-letter excerpts from five US states. What the four-pass pipeline actually writes.
For personal injury and workers’ comp principals, managing partners, and litigation operators evaluating reasoning-grade AI for their firm. Solo and small practices start a trial themselves; sales-assisted for mid-size and enterprise.