I build LLM-powered agentic AI for in-car assistants, based in the Netherlands, with 9+ years of shipping ML systems end-to-end.
- Agentic synthetic data: LLM-driven training-data generation with per-stage subagents, tool simulation blended with real API calls, and LLM-as-judge quality loops
- Multi-agent simulation: event-driven agents with personas and private memory, running against live Microsoft 365 via MCP to produce fully traceable training data
- Serving and fine-tuning: self-hosted vLLM with per-stage model routing (local Mistral / Gemma plus Azure OpenAI), LoRA fine-tuning for router, tool-call and multi-turn behavior
- Applied NLP: intent classification on real in-car dialog, lifted 60% to 85% by LoRA fine-tuning Llama 3
- Senior Machine Learning Scientist (automotive AI assistants)
- Previously Senior Data Scientist at Shopee: pricing and product matching at 13M+ product scale
- MSc Applied Data Science, Utrecht University
- Invited keynote speaker on agentic AI, CTC AI Tech Conference 2026