The whole thing, in order.

AI Engineer

Chiron Consulting · Remote

Marketing consultancy. Freelance, then my M1 internship (May to Aug 2026), then a contract to Dec 2026. I built the keyword-retrieval system the content team uses every day to draft posts in each client's voice. Retrieval raised the automatic voice score in testing, but the score didn't match blind human ratings when I checked. I also built the team's Claude Code workspace for 16 client brands, and onboarded the team, who now run it day to day.

AI Security & Governance Consultant

ACE-NSI Cyber Security · Part-time, remote

Cybersecurity consultancy selling AI governance to regulated firms. Designed a Shadow AI assessment that risk-scores staff AI use and maps it to the EU AI Act, GDPR, NIS2 and DORA. Tested two AI-security platforms for prompt data-loss interception and agent/MCP risk scoring, and wrote the vendor selection guide. Found in Microsoft's own docs that Purview blocks prompts for only about 18 AI apps, in Edge only, not Claude or Mistral.

AI Engineer

Freelance · Paris, Remote

Set up Claude Code for small businesses and an individual sales leader, each with a working folder it reads when it opens and skills for the jobs they repeat.

Student Mentor & Sports Lead

Charles University, Prague · Prague, Czechia

Guided incoming international computer science students and ran sport for the maths and physics students.

MSc Artificial Intelligence

Université Paris-Saclay, LISN · Paris, France

M1 finished in June 2026 with an overall mark of 15.097/20. I'm in the M2 year now: retrieval, evaluation, NLP, neuro-symbolic methods and generative models, then the internship, which is the whole second semester.

BSc Computer Science with specialisation in Artificial Intelligence

Charles University, Prague · Prague, Czechia

Final exam graded 1 (Excellent) overall. Where I learned to code, and where I got into machine learning. Algorithms, software engineering and lots and lots of maths, with the artificial intelligence specialisation on top. My thesis, a terrain generator built for the game Engine Evolution (1.5M+ downloads), was graded 1 and extended into the IEEE paper below.

Business Science

University of Cape Town · Cape Town, South Africa

A year of business science: economics, accounting and statistics.

Constraint-Aware Terrain Completion from Sparse Fixed Regions for Game Heightmaps

Petr Šimůnek, Oliver Wakeford

IEEE Conference on Games 2026 (CoG), Madrid

Full paper, IEEE CoG 2026 (oral). Grew out of my BSc thesis, defended 2025. It keeps the designer's fixed heights exactly: mean-height bias about 81% lower than the blurred production baseline (0.01346 against 0.07112), at comparable runtime, and 2.1× to 23.9× faster than the harmonic baseline.

Second author. The work started as my BSc thesis, and my supervisor extended the method, ran the experiments and led the writing.

Inverse distance weighting scores 0.01351 on the same metric, a margin of about 0.4%. And total variation is worse, 933.5, against 480.9 for the harmonic baseline, so the surface is rougher.

Does retrieval make generated copy sound like the client?

Oliver Wakeford

Master's internship report, ACL format

Submitted August 2026. Not peer-reviewed, and not published

An internship deliverable assessed by the university, not a publication. The corpus belongs to a client and can't be shared.