what i've been doing
- 2025–agent infrastructure and evaluation harnesses · friends of cartel
- 2026–evals on inspect: what agents write where they think no human reads · independent research
- 2026–wtf, ai?: an open evening in leuven, and a dossier in dutch of what ai is actually doing · wtf, ai?
- 2026ai safety and x-risk communications · plzdontkillus
- 2025–kaios: llm integration and generative sound · kotopia
- 2024–25autonomous agents and image-training pipelines for inku · inect
- 2022–24advanced master in ai, speech and language technology · ku leuven
- 2022–23intent classification for citizen questions · the belgian government
- 2021–22master in digital text analysis · university of antwerp
- 2021research communications and websites for muhai and beehaif · ai lab, university of brussels
- 2018–21bsc applied linguistics · ku leuven
repositories
- 2026evalsi give an agent a diary and tell it nobody will ever read it. then i read it
- 2026voting-eval-critiqueapart research built a judge to grade how models answer questions about voting. i graded the judge
- 2026plzdontkillus auditeven the people trying to save us leave doors open. i found lightcone's, one of them critical, and handed every key back
- 2025mcp-server-starterevery mcp server i've shipped grew from these bones. take them
- 2026claude-voicei talk, claude code listens, and nobody has to type
- 2026iphone-ereaderan old iphone from my drawer, turned into an e-reader that has forgotten how to be anything else
- 2025–agent infrastructure · privatetwenty-something mcp servers that agencies wake up to every morning, each one doing a whole job start to finish
- 2025chatgpt connectors · privatei taught chatgpt to use our tools, then taught claude the same trick
- 2022pythiamy master thesis, back in 2022, before ai ethics went mainstream and before i had really dived into safety. i built a machine that judges morality, then made it show its work