Layer04
Operating Systems
ArthurAI™ — Agentic Learning OS
VLE, CLE, SLE, and ULE — productized vertical platforms with structured-workflow UI, embedded reasoning, and compliance-grade audit trails. Customers interact here.
TechnologyEve-Education™ F5/reasoner
Eve-Education™ Fusion v5.
A fusion of five cooperating reasoning models in a compositional fabric — purpose-built for educational environments. Deployed on Eve-Grid™. Trained on Eve-Genesis (Education Edition) — synthetic reasoning data, not customer data.
Three deep-dives
The reasoning core, the synthetic corpus, the cloud substrate.
The reasoning core
F5/reasoner →
The Fusion of five cooperating reasoning models. Phi-3 classifier, Phi-4 SRM, and three frontier slots. Each layer named, each role explained.
The synthetic corpusEve-Genesis →
Eve-Genesis (Education Edition) is the synthetic educational reasoning corpus that LoRA-fine-tunes the Phi-4 educational reasoner. Pedagogical taxonomies, curriculum-aligned reasoning traces, no customer data ever used for training.
The cloud substrateEve-Grid →
The Azure-native architecture that hosts F5/reasoner. HYVE = Hyper Yielding Vectorized Entity. Multi-tenant isolation, regional data residency where regulatorily required.
Four layers, top downSection 02
How ArthurAI is composed.
Layer03
Digital Employees
Arthur — the educational-reasoning specialist
A workflow-bound AI specialist with persistent identity, continuous learning, and Azure AI Search-backed vector memory. Arthur reasons; the educator decides.
Layer02
Compound Reasoning Models
Eve-Education™ Fusion v5
Five cooperating models — a Microsoft Phi-3 classifier, the Phi-4 educational reasoner fine-tuned on Eve-Genesis (Education Edition), and three frontier models as consultants, composed best-fit per request.
Layer01
The Substrate
Eve-Grid™ · Eve-Genesis™ · Eve
Eve-Grid™ on Microsoft Azure hosts the fabric; Eve-Genesis (Education Edition) fine-tunes the reasoner via LoRA; Eve is the meta-orchestrator. The body, the cognition, the training.
A compositional fabric, not a stack
What this architecture buys us.
Per-agent specialization without per-agent rebuild
Each Digital Employee gets the appropriate set of frontier and proprietary models for its workflow, without forking the underlying architecture.
Modality awareness
Vision tasks (diagram analysis, handwritten work review) route to vision-capable models. Text reasoning routes to text-strong models. Cost-sensitive tasks route to small fast models.
Regulatory flexibility
Where an institution or jurisdiction prohibits a specific model provider, that provider is swapped out of the relevant agent's frontier slot without rebuilding the agent.
Frontier-release future-proofing
When new frontier models are released, they are integrated into the appropriate slots without retraining the proprietary classifier or Small Reasoning Model. The architecture absorbs frontier progress.
Auditability by design
Compliance engineered into the reasoning substrate.
Every output from a Digital Employee is auditable and traceable to its underlying reasoning. The Operating Systems are structured-workflow, not free-form conversational, by design — because free-form interfaces are incompatible with the auditability requirements of regulated industries.
ArthurAI™ produces structured analytical output with evidence presented openly, reasoning visible at every step, confidence calibrated, every source cited. The educator decides what to act on. The AI reasons; the educator decides.
A note to the reader
Talk to our architecture team.
For CIOs, CTOs, procurement teams, and partner integration teams evaluating architecture fit.
Start a conversation →