After 10+ years in AI/ML — a PhD at KU Leuven, publications at ICML, ICCV, and ICDAR, and building production AI systems at companies like Instabase and Contract.fit — I've founded my own consultancy BV based out of Belgium.
I'll still be engaged with Instabase as a GenAI Tech Lead, building best-in-the-biz Agentic Architectures for (Cross-)Document applications — yet I'm opening myself up to the world to discover what other problems I can contribute to.
The company is called Probably Approximately Human — a nod to Probably Approximately Correct (PAC) learning, a theoretical ML framework I adored when starting out — the foundation that tells us when and why machines can actually learn. Swap "Correct" for "Human" and you get what I'm after: AI systems that are good enough to be useful, grounded enough to be trusted, and honest about what they don't know. That's the kind of AI I build.
And the approximately human part? Full transparency: through recent innovations in general-purpose coding agents, I've built a specialized AI-augmented setup that lets me produce the output of a small team. You're not just hiring one person — you're hiring one person with a force multiplier.
I'm open for business. Here's who I'm looking to help.
You're a CTO or VP Engineering and your team is struggling to keep up with GenAI.
The landscape moves weekly. Your engineers are brilliant, but nobody has time to separate hype from what actually ships. I help you build a clear AI roadmap: where LLMs, agentic systems, and RAG architectures genuinely move the needle — and where they don't. From strategy to working prototypes, I bridge the gap between "we should do something with AI" and production-grade solutions that deliver.
Your organization is drowning in documents.
Invoices, contracts, reports, forms — terabytes of unstructured data that someone is still processing manually. Document AI is literally what I did my PhD on. I built the DUDE benchmark (adopted at ICCV 2023), led document understanding systems at Contract.fit for 7 years, and designed agentic automation pipelines that hit 100% accuracy through self-correction. If you have a document problem, I've probably already solved a version of it.
You're an academic PI or R&D lead writing a grant proposal and need a second pair of specialized eyes.
I've been on both sides of VLAIO grants — as a Baekeland fellow and as a co-author on multiple HBC innovation grants. I know how to frame AI/ML research so it lands with reviewers: technical depth that's credible, impact narratives that resonate, and workplans that are actually achievable. Whether it's VLAIO, Horizon Europe, or other EU funding instruments, I can sharpen the AI/ML sections of your proposal or help you build the technical case from scratch.
"Why should I pick you?"
Don't take my word for it — ask around. My closest current and former teammates (several of whom have since moved to Databricks, Anthropic, OpenAI, and Google's Gemini team) consistently said the same thing in peer reviews: most knowledgeable person in the company on LLMs. I don't need to collect testimonials. The people who've worked with me know. And I thrive on prototypes — give me a hard problem and a week, and I'll hand you back something that works.
What I bring to the table
- PhD in Computer Science (KU Leuven) + 15+ peer-reviewed publications
- Hands-on experience shipping AI at Instabase, Contract.fit, Snowflake, Hugging Face, Oracle
- Research collaborations with Oxford, TU Darmstadt, CVC Barcelona, UCSD, UNC Chapel Hill
- Deep expertise in Document AI, Agentic Systems, LLM/VLM engineering, and evaluation
- 6 languages (Dutch, English, Spanish, French, German, Portuguese) — available globally from the EU
Most of my career has been AI close to the enterprise — but honestly, I'm open to any problem where you think a skillset in approximation could help. Healthcare, climate, logistics, creative tools — if it's hard and interesting, I want to hear about it.
If any of this resonates — or if you know someone it might help — I'd love to connect.