AT WORK · INTEGRATION TEAM
Integration-team chatbot
Built for the team I work in. Significantly reduced the time it takes to identify integration issues.
FORWARD DEPLOYED ENGINEER · COPENHAGEN
Proof before promises
I embed with your team and deliver working software on your real data: demo-first, scoped honestly, handed off clean.
FLAGSHIP · SOLO-BUILT
A sensor-agnostic predictive-maintenance platform: an ML perception pipeline plus Claude agents that report deviations in plain language. It trains on healthy operation only: no failure labels, no per-asset model engineering.
“My vibration sense has been increasingly surprised for the last 40 hours. Thermal and pressure feel normal.”
| DATASET | RESULT | SEPARATION |
|---|---|---|
| Turbofan · NASA C-MAPSS FD00130 UNSEEN ENGINES | ρ = 0.762 · 100% engines positive | AUC 0.996 |
| Turbofan · FD002, 6 regimes78 UNSEEN ENGINES, THE HARD SUBSET | ρ = 0.615 · 100% engines positive | AUC 0.95 |
| Bearing vibration · NASA IMSRUN-TO-FAILURE | flagged 75 h and ~470 h before end of run | AUC 1.0 / 0.96 |
| Pump audio · MIMII / DCASE 20204 REAL PUMPS, FACTORY NOISE | beats the purpose-built audio baseline on 2 of 4 machines | AUC 0.687 vs 0.715 |
Honest status: research-grade pipeline, validated end-to-end on real hardware and public benchmarks. Not yet productized. The per-sense triage reproduced known bearing physics unsupervised: the spectral sense fired first, up to 19 days before broadband energy moved.
AT WORK · INTEGRATION TEAM
Built for the team I work in. Significantly reduced the time it takes to identify integration issues.
PERSONAL PROJECT · LIVE
A live site that evaluates Greek property, built end-to-end as a personal project.
ΦΡΙΞΟΣ · PHRIXUS
A golden ram flew Phrixus out of a death sentence and across the sea. The story gives my work its shape: I get sent across, embed where the work is, and leave behind something worth keeping. The sea is the AI age now and every business has a crossing to make.
THE RAM BECAME ARIES · HORNS: TRUE GOLDEN RATIO SPIRALS, ×φ PER QUARTER TURN
THE MEASURE OF A HANDOVERWhat the ram left behind was worth a voyage of its own. That is the bar I hold a handover to. The image is an AI-generated study, the only thing on this page that is.
Forward deployed means the work happens inside your context, not mine: your data, your workflow, your definition of done. The procedure is the same at any scale.
I start by listening for how the work actually happens (the real workflow, not the org-chart version) and what solving the problem is worth before anything gets built.
If I don't believe I can make a measurable difference to your business, I say so at the door. The projects I turn down are as load-bearing for trust as the ones I take.
The narrowest piece that runs end to end on your real data and proves the value. Not a platform, not a roadmap. A proof.
We agree what “working” means as concrete test cases before I build. Then: build, demo, report the pass rate. Nothing stays unshown for a week.
Documentation, a walkthrough, and a follow-up check. The system becomes yours, not a dependency on me.
EXPERIENCE
Integration Specialist: enterprise integration architecture and end-to-end data flow.
Integration Specialist.
Building agentic systems across personal projects, Sentys among them.
EDUCATION
NLP and computer-vision machine learning, from back when nobody called it AI.
Mixed hardware and software, comfortable the whole way down the stack.
Generalist by design: AI solutions, integration architecture, end-to-end data flow. Easy to talk to, and I present well: to engineers and to the people who sign.
F. ANDREOU55.676° N · CPH
A slow workflow, a data stream nobody reads, an AI idea that needs an honest feasibility answer. I'll tell you plainly whether I can help, and show you working proof if I can.
COPENHAGEN-BASED · WORKS REMOTE ACROSS EUROPE