Research

Governance-first research for enterprise AI.

Our work focuses on deterministic orchestration, human authority, and why auditability and data ownership must be architectural.

The three paradigms of AI systems.

Probabilistic tools are insufficient when consequences are regulated and irreversible.

Probabilistic assistants

Helpful, but unverifiable. Output variance creates compliance and audit exposure.

Agentic workflows

Autonomy without governance scales risk faster than value.

Deterministic orchestration

Explicit state, policy enforcement, and human approval at every step.

Why current approaches fail.

Protocols solve connectivity. They do not solve governance.

Audit gaps

Without deterministic workflow, you cannot prove why a decision happened.

Policy drift

Compliance rules change faster than model prompts can keep up.

Human authority loss

Agentic systems skip approvals by design, not by exception.

The orchestration layer: why MCP needs governance.

MCP, AgentKit, and ADK standardize connectivity. They do not answer who controls it.

The protocol explosion

Anthropic MCP connects models to tools and data.
OpenAI AgentKit defines deterministic agent patterns.
Google ADK offers agent tooling on Google Cloud.

The governance gap

Is this user allowed to make this specific request?
Does this operation comply with policy and regulation?
What audit trail documents this decision?

Human AI as MCP orchestrator.

We sit above protocols and enforce governance before, during, and after execution.

Before

Policy checks, authorization, and workflow eligibility.

During

Model routing, validation contracts, and deterministic task control.

After

Audit records, state transitions, and human approvals.

+----------------------------------------------+
|              Human AI Orchestration          |
|  +----------------------------------------+  |
|  | Deterministic Control Plane            |  |
|  | (FSM/Petri Nets, Policy Enforcement)   |  |
|  +----------------------------------------+  |
|                    |                         |
|  +----------------------------------------+  |
|  | Protocol Adapters                     |  |
|  | MCP | A2A | ACP | REST/gRPC            |  |
|  +----------------------------------------+  |
|                    |                         |
|  +----------------------------------------+  |
|  | AI Models (Claude, GPT, Gemini, etc.)  |  |
|  +----------------------------------------+  |
+----------------------------------------------+

Nebari and the Open Teams ecosystem.

A natural extension of the Open Teams partnership and the scientific Python stack.

Nebari foundation

Kubernetes-native infrastructure for JupyterHub, Dask, and ML workflows across clouds or on-prem.

Human AI addition

Govern ML workflows with deterministic policy, audit trails, and multi-model routing.

Strategic alignment

The combined stack delivers enterprise AI operations with neutral governance.