I'm an IT infrastructure and security specialist with 15 years in network administration and security engineering.

I also design, build and run production web applications end to end — I write them, deploy them, and keep them working for real users.

I build under Simpliflex Systems Ltd.

Education software that works with no internet connection, built for real learners.

Built in: per-child accounts · parent dashboards · printable worksheets · offline text-to-speech · subscription billing

Probabilistic match-outcome models across three sports. The interesting part is not the prediction — it is proving the confidence attached to it is honest.

How they are evaluated

• Calibration over accuracy. A model that says "70%" must be right about 70% of the time.
• Scored on Brier score, log-loss and expected calibration error, not on hit rate alone.
• Isotonic recalibration where raw confidence drifts. On the basketball engine this cut expected calibration error from 0.050 to 0.009 across 325 leagues.
• Walk-forward backtesting, so a model is only ever judged on matches it could not have seen.

🔍 I test what the user actually touches. A content check proving a question has a valid answer key tells you nothing about whether the screen can draw it. I learned that the hard way, and now gate the render path too.

🧪 Every new check gets proved. I revert the fix and confirm the test fails before I trust it. A gate that cannot fail is not a gate.

🚦 Automated checks before every ship. Simpliflex Learn runs 26 of them: content validation, navigation, authentication, marking, accessibility and read-aloud coverage.

👥 Real users set the priorities. The last release came straight from a learner's message about her quiz skipping questions.

Building with

Currently deepening — Kubernetes · Terraform · AWS architecture · applied cybersecurity in a home lab

Happy to talk network and security engineering, education software, or shipping small products solo.
_{Pronouns: He/Him}