Everything I know how to do, arranged as a room. Walk through it, and you will know me by the objects I keep.
Forecasting demand across more than 15,000 locations, in 40+ countries and 20+ languages, for Fortune 500 operators. The work only counts if it holds at that scale, in production, after the demo is over.
Retrieval pipelines, cross-encoder rerankers, autonomous agents, and the evals that keep them trustworthy. Guardrails and measurement are not an afterthought, they are the product.
A daily "Today in AI" field note: one story, gone deep, from a production-builder's seat. What shipped, what it means, what to ignore. The discipline of writing it down is how the intuition compounds.
Read the writing →An aerospace M.Sc. from TU Delft, then technical product ownership and PMP-certified delivery. The judgment to pick the right problem and carry it from prototype to something a business can defend.
Trace the path ↓RAG pipelines, autonomous agents, and forecasting platforms across 15,000+ locations in 40+ countries. I turn ambiguous problems into measurable systems, at Fortune 500 scale.
Retrieval, ranking, agents, evals, and the discipline of keeping them healthy after the demo is over.
I build production AI systems end to end. Most of my recent work is Fortune 500 deployments where the signal lives in messy enterprise data and the buyer is a non-technical executive who needs a number they can defend.
My path to AI started in aerospace engineering at TU Delft, with a thesis on neural networks for turbulent airflow simulation. That mathematical rigor became my edge. Today I work across RAG pipelines, agentic workflows, evaluation harnesses, and the production observability that turns a clever notebook into a system you can trust.
M.Sc. Thesis · TU Delft A Neural-Network-Coupled Variational Multiscale Method for 3D Turbulent Channel Flow Computational Fluid Dynamics · Deep Learning · Numerical Methods →End-to-end retrieval for a Fortune 500 QSR chain: semantic search across thousands of locations, custom cross-encoder reranking, NeMo Guardrails. 20+ languages, sub-5s latency.
Ensemble forecasting (XGBoost + LSTM) across 15,000 to 20,000 locations globally. +8 to 12% accuracy, -6 to 8% stockouts, automated triggering of thousands of campaigns monthly.
LLM-based operations agent for a Fortune 500 cruise-line operator. Classifies 50 to 80 daily signals and recommends action with sandboxed execution and human-in-the-loop governance. -60% resolution time.
New opportunities, collaborations, or a chat about AI, evals, or turbulence models. Reach out.
Say hello