emre yavuz
ships ai in places where it has to behave.
welcome. choose a section.
↑↓ or click · enter to open · 1–6 direct · ? help · t theme · esc back
$ work
research
products
- $ ask
- aria-live.app
- growbuddy.ai
- more coming.
enterprise services
- ai governance
- mlsecops
- ai strategy
- ai change management
$ whoami
i've spent more than a decade building, supporting, and breaking software where the cost of wrong is public.
now, i ship ai in swiss regulated finance and design the systems it runs on.
focused on post-training lately: i train, break, and measure language models and the reward signals that steer them, and write down what actually happened.
the projects on this page, the research hub, and the repos on github and hugging face show the work.
deliberate architecture proactive safety audit-ready defaults
$ how
at work
- compliance and ship are the same problem.
- strategy is a technical decision.
- the deck is not the system.
- proactive safety, not reactive.
- governance as foundation, not compliance tax.
- privacy-by-design, or not at all.
- analytical by default. strategic by habit. unreasonably thorough.
in life
- evidence where it counts. virtue where it doesn't.
- good math isn't the same as good.
- ultimately, it comes down to taste.
- SNR = 10·log₁₀(P_signal / P_noise) = 10·log₁₀(4) ≈ 6.02.
- labels after the practice, not before.
$ notes
empty for now. writing lands here.
queued: governance that holds up · MLSecOps in practice · cost-aware ai.