About

Leone Lage Perdigão

AI Architecture & Delivery · AI Security · Governance & Evaluation

Path

I have worked in software engineering since 2010. Through the 2010s I moved into cloud architecture and platform engineering, as cloud changed how software is built. I moved into AI because it is the next change of that size, and I want my expertise where the industry is heading. I took an MSc in Artificial Intelligence at the University of Bath, with distinction, while leading AI engineering teams.

TaoQ AI is my independent practice. I design, build and secure AI systems with engineering teams at European tech companies, so they hold up in production and in an audit. I am an ISO/IEC 42001 certified Senior Lead Implementer (PECB).

The work is concrete: threat models, agent red-teaming, EU AI Act gap analysis, and the technical documentation the Act requires. I stay in the code with the team rather than hand over a report.

From October 2024 to August 2026 I was Lead AI Architect in PostNL's AI Centre of Excellence. I led the work on its AI reference architecture, covering machine learning, generative and agentic AI, and AI at the edge. It became the standard across a 30,000-person organisation and more than 100 use cases.

I led three teams there, from prototyping through to company-wide adoption and governance. The MLOps and LLMOps standards were built to meet the EU AI Act. The aim then is the aim now: AI you can defend in an audit.

Before that I was an Advisory AI Engineer at IBM, the technical authority for AI solutions across Northern, Central and Eastern Europe. I architected enterprise systems on WatsonX and led the AI side of IBM QRadar EDR, IBM's endpoint security product.

From 2021 to 2023 I was Lead Software Engineer in PostNL's Cloud Centre of Excellence. I architected its serverless-first AWS platform, which served 100+ teams across 1,000+ AWS accounts.

I mentor engineers through IBM's Blue Core programme and on MentorCruise.

Open source

I created Ziran, an open-source tool that tests AI agents for security flaws. It looks for dangerous combinations of tools that single-prompt tests miss. I also contribute to NVIDIA Garak and Hugging Face Transformers. Building tools keeps my architecture decisions honest.