The compositional fabric, in five slots.
Eve-Legal™ Fusion v5 composes five cooperating reasoning models per request. The explorer below is the architecture as we ship it — every slot, every model, every reason that model occupies that slot. Hover or click to drill in. The data binds to the same registry the demand-letter pipeline uses; what you see is what runs.
Five cooperating models, one reasoning loop.
Hover or click any slot below to see the model that occupies it, the role it plays in the reasoning loop, and why that specific model is in that specific slot. The architecture is plug-and-play: providers swap, frontier slots compose per request, the legal reasoner stays.
Microsoft Phi-3
Classifier
Sub-200ms routing across workload type, modality, reasoning intensity, and context-length needs.
Why this model occupies this slot. Lightweight 3.8B-parameter classifier. The first responder of the stack — decides which downstream models compose for the matter at hand.
Phi-3 routes every inbound request to the appropriate downstream slot in under 200 milliseconds. Workload type, modality, reasoning intensity, and context-length need all enter the routing decision.
The supervisor sits one layer up.
Above the five cooperating models, Justine — the named Digital Employee — runs the supervisor + sub-agent pattern. Justine supervises stage-specialized sub-agents (intake, medical, valuation, strategy) across the case lifecycle. The sub-agents are not separately branded. This is the architecture that makes Mass Tort possible: one Justine, thousands of plaintiff sub-agents, one reasoning context.
F5/reasoner is the middle. The architecture is three tiers.
Justine + named agents
The orchestration layer. Justine coordinates a roster of stage-specialized agents — Intake, Medical, Valuation, Strategy, Deposition, and the Case Advisor. The supervisor pattern dispatches each sub-task to the right agent, with the right F5 composition. The sub-agents are not separately branded — Justine is the only Digital Employee with a name.
Five cooperating models
The compositional fabric. Classifier · legal SRM · frontier slots · long-context slot — composed per request. Plug-and-play, provider-swappable, matter-aware in composition. This is the tier the explorer above shows in depth.
Curation Layer (9 modules)
The input layer. Before the matter reaches F5, nine modules clean the record: OCR Janitor, Provider Canonicalizer, ICD-10/CPT Code Validator, Incident-Date Splitter, Damages Authority Resolver, Encounter Coalescer, Lien Joiner, Composer Migration, and a Curation Report surfaced in the attorney’s Review Panel. The AI reasons on clean data because the data is cleaned first.
Every legal-AI vendor talks about their model. The honest picture is three tiers — one to clean the inputs, one to reason, one to orchestrate. Each tier has its own engineering. F5/reasoner is the centre of the architecture; the tiers above and below are what make F5 effective on legal work.
See JustineAI™ in your practice.
For PI principals, managing partners, and litigation operators evaluating reasoning-grade AI for their firm. Self-serve trial available for solo and small practices; sales-assisted for mid-size and enterprise.