Architect · Builder · Leader
VP, Cloud & Platform Architecture at Deutsche Bank. I work on two connected problems: the management systems regulated enterprises need to scale AI responsibly, and the protocol infrastructure that makes multi-agent AI verifiable. The first is an operating model and governance challenge. The second is an engineering one. In regulated environments, you have to solve both.
Flagship Work
The management system for enterprise AI.
Why 95% of enterprise AI programs fail to deliver P&L impact, and what the 5% do differently. Five operating principles, four maturity stages, and the governance, architecture, and measurement frameworks that convert AI capability into business value. Includes an interactive AI Readiness Assessment with shareable scorecard.
Read the playbook The framework Take the assessmentData platform architecture for regulated enterprises. Mesh patterns, quality frameworks, lakehouse design, and the operating model that makes data AI-ready.
ReadResearch
The technical foundation beneath enterprise AI governance.
When AI scales to multi-agent architectures, governance needs protocol-level answers. Published research on identity, delegation, provenance, and collective reasoning, with experiment code and open-source implementations.
~2,000 MCP servers scanned, zero authentication. AIP adds verifiable delegation tokens with 0.049ms verification and 100% adversarial rejection. Python SDK on PyPI, framework adapters for CrewAI, Google ADK, and LangChain.
Project hub arXiv Blog post Code~12x latency reduction on easy tasks, 37% fewer tokens, 96% attack detection. Six research questions with honest null results.
Interactive paper arXiv Project page+0.95 over debate on non-routine tasks. 9.56 on hidden-profile (best in study). Debate wins on routine tasks. 62x cost not justified.
Interactive paper arXivSelected Writing
Enterprise AI strategy, governance, and systems architecture.
The gap between AI strategy documents and execution systems. Why the organizational share of the problem outweighs the technical one.
ReadWhy traditional model risk management fails for LLMs. Hallucination policy, GenAI risk dimensions, deployment gates, and what a complete governance framework looks like.
ReadAI programs fail at the middle layer. The people who should be supervising AI-augmented workflows were never equipped to do it.
ReadOpen Source
Selected infrastructure projects.
RAG patterns benchmarked for enterprise. Five chunking strategies, five retrieval patterns, evaluation harness, guardrails, and observability.
AI governance for regulated industries. Risk classification, model lifecycle controls, EU AI Act mapping, and worked examples for credit scoring and fraud detection.
Background
Contact
Open to leadership conversations.
If you're shaping the next stage of enterprise AI capability, data platform modernization, or technology transformation — I'd welcome the conversation.