Problem
Name the real constraint — the one everyone else designed around instead of through.
I'm Konstantinos Zervakis — I build production ML, LLMs & multi-agent systems for environments where a wrong answer costs the vintage, a wildfire response, or a satellite mission. Everything here follows how I actually think: the problem, the solution, the result.
A benchmark impresses for a minute. A solved problem holds up when a wrong call risks the vintage and a missed defect ends a satellite mission. So I don't lead with model names — I lead with the constraint that mattered, the design that removed it, and the number that proves it worked.
I'm relentlessly curious — first to try what just shipped — and I aim high on purpose. But curiosity without rigor is just noise, so every experiment lands as something measurable. The work below isn't a gallery. It's a stack of case files.
Now Software Test Engineer @ OHB Hellas (aerospace, ECSS V&V/PA) · PhD candidate, autonomous UAV navigation @ Hellenic Army Academy
Name the real constraint — the one everyone else designed around instead of through.
The smallest system that removes the constraint for good. Typed, tested, and honest about its limits.
A number. Accuracy, latency, F1, cost, hours saved. If it can't be measured, it isn't done.
Each one, the same shape: the problem, the solution, the result — with a number attached.
Earlier case files — smaller scope, same shape.
Agent framework of choice: Pydantic AI — typed contracts, graphs, one-layer retries, offline-testable.
Pulled live from the GitHub API — the fastest-rising repositories of the last few days. Not a screenshot. It's genuinely current every time this page loads.
api.github.com/search
What sharpens the way I think, on and off the keyboard.
The good ones don't wait. Based in Crete — working worldwide.
zervakisai@gmail.com →