My Research
Current Doctoral Research: The High-Risk AI Evidence Matrix
My primary doctoral research utilizes applied Design Science Research (DSR) to resolve the "evidentiary gap" between qualitative AI regulatory mandates—such as the EU AI Act and the NIST AI RMF—and execution-layer MLOps engineering. To combat the "Compliance Illusion" inherent in purely process-driven standards like ISO/IEC 42001, I architected the High-Risk AI Evidence Matrix and the Rights-Based QMS Overlay.
This socio-technical framework translates abstract statutory and human-rights requirements into cryptographically verifiable JSON-LD telemetry, disparate impact metrics, and immutable Human-in-the-Loop (HITL) override logs. By mapping these machine-readable artifacts directly to compliance controls, the research provides General Counsels, Notified Bodies, and engineering teams with the exact objective, Provided-By-Client (PBC) evidence necessary to secure Continuous Authority to Operate (cATO) and European market authorization (CE marking).
Forthcoming Research Initiatives
Telemetry Schemas for Autonomous Multi-Agent Workflows
As organizations move beyond single-model deployments into agentic AI, traditional point-in-time auditing fails. This project develops standardized telemetry payloads specifically designed to track cascading decision logic and automation bias across multi-agent systems. The research aims to define how independent AI agents can cryptographically sign their outputs and handoffs, creating a verifiable chain of custody that satisfies Article 17 logging requirements under the EU AI Act even when humans are entirely removed from the execution loop.
Cryptographic Liability Severance for Downstream AI Integrators
As the enterprise adoption of General Purpose AI (GPAI) scales, a critical legal friction point has emerged between foundational model providers and downstream integrators. When an AI incident or regulatory breach occurs, determining fault within this shared-responsibility ecosystem is currently technically opaque. This project investigates mechanisms to technically and legally sever a downstream integrator’s liability from the underlying foundational model's "black box." By architecting immutable input/output boundary telemetry, API payload hashing, and localized prompt-guard logs at the integration layer, this research aims to provide downstream deployers with the exact cryptographic evidence required to definitively isolate fault. Ultimately, this framework ensures that enterprise integrators can legally defend their systems and are not held liable for upstream foundational model drift or latent algorithmic biases.
Dynamic Regulatory Ontology Mapping (DROM)
With global AI regulations fragmenting rapidly across jurisdictions, multinational corporations struggle to maintain unified compliance architectures. This project focuses on building a dynamic, machine-readable ontology that cross-walks technical requirements between the EU AI Act, the U.S. AI Bill of Rights, NIST frameworks, and emerging APAC regulations. The goal is to engineer a unified algorithmic control inheritance model, allowing MLOps teams to write compliance telemetry once and automatically map the evidence to multiple global statutory frameworks.