About
The short version, told straight.
I'm Oliver. I install AI inside small businesses, and I build the tools I use to do it. I live in Paris and I am halfway through a master's in AI and data science at Université Paris-Saclay.
Before that I read Computer Science at Charles University in Prague, specialising in machine learning, and graduated with First-Class Honours. My thesis was on procedural terrain generation with fixed points. No machine learning in it at all, as it turns out. The method is constraint propagation: it pushes slope constraints out from the designer's fixed anchors into upper and lower height envelopes using slope-weighted path costs, then fills the gaps and clamps to the envelope so the anchors stay exact. My supervisor extended it into a paper, and it was accepted for oral presentation at IEEE Conference on Games 2026. I am the second author.
I started consulting while studying. First a marketing agency, where I built a retrieval-augmented content system and then spent longer measuring whether the retrieval had done anything than I did building it, which is how I found out that one of the numbers I had already reported was wrong. Then a cybersecurity consultancy, on AI governance and Shadow AI exposure. Then five installs, which is the work I want more of.
The through-line is that I would rather know why something works. Most people selling AI consulting learned the tools. I was taught what is underneath them. Most of the time that changes nothing at all.
Outside work: the gym, running along the Seine, and trying to get my French past B1. Slower going than I'd like.

Experience & education
Full CV, with publicationsPre-Sales AI Security Consultant
Cybersecurity consultancy · Remote
Designed a Shadow AI exposure assessment methodology, and built the research and outreach pipeline that feeds it. Consulting on AI governance and on GDPR, EU AI Act and NIS2 compliance for organisations in regulated industries.
2026 – Present
MSc AI & Data Science
Université Paris-Saclay · Paris, France
Machine learning, deep learning, information theory, information retrieval and data engineering, taught in English. M1 completed June 2026 with an overall mark of 15.097/20; M2 runs to June 2027.
2025 – 2027
Freelance AI Consultant
Independent · Paris, Remote
I install Claude inside small businesses: one folder on the client's own machine that their AI reads and writes, built around how the business actually runs. Five completed engagements, ranging from a company two weeks old to one carrying years of files and no system that had ever stuck.
2025 – Present
AI Consultant
Marketing agency · Remote
Built a retrieval-augmented content system and the application that runs it, then measured whether the retrieval actually worked. Written up as a master's internship report, including a result I had to retract.
2025 – Present
Teaching Assistant
Charles University, Prague · Prague, Czechia
Ran lab sessions and marked coursework for undergraduate computer science. Explaining the same idea to twenty people who each misunderstood it differently is the closest thing to consulting practice I have had.
Sept 2024 – June 2025
BSc Computer Science with specialisation in Artificial Intelligence (First-Class Honours)
Charles University, Prague · Prague, Czechia
Where I learned to code, and where I got into machine learning in 2022, a few months before ChatGPT came out. Algorithms, software engineering and mathematical logic, with the artificial intelligence specialisation on top. I'd done a year of business science before this and preferred this by a distance. I also spent an academic year running the weekly lab sessions for an undergraduate course here, about twenty students a time, and it was the first time I'd had to explain any of this to people who didn't already get it. The thesis, on procedural terrain generation, became the IEEE paper below. I won two public-speaking competitions while I was there.
2021 – 2025
Business Science
University of Cape Town · Cape Town, South Africa
A year of business science (statistics, economics and finance) before moving to Prague for computer science. It's why the commercial side of this work isn't something I had to learn later.
2020 – 2021
Skills & tools
Graded, because a flat list tells you nothing. Core means I would take a hard problem in it. Familiar means I have shipped something and would need a day to warm up.
AI & machine learning
- Retrieval & RAGCore
- Model evaluationCore
- LLM application designCore
- Deep learning (PyTorch)Working
- Fine-tuning (LoRA / PEFT)Working
- Generative modelsWorking
- AI governance (GDPR, EU AI Act, NIS2)Working
Development
- PythonCore
- TypeScriptCore
- Next.js / ReactCore
- SQLite / PostgreSQLWorking
- Flask / FastAPIWorking
- Unity / C#Familiar
Tools & platforms
- Claude API & Claude CodeCore
- GitWorking
- VercelWorking
- SupabaseWorking
- Tailwind CSSWorking
- JupyterFamiliar
Consulting & business
- Technical writingCore
- Client-facing deliveryCore
- AI strategyWorking
- Pre-salesWorking
- Public speakingWorking
Languages
- EnglishNative
- FrenchB1, working on it