Skip to content

AI systems rarely
fail loudly

They return fluent, well-formatted answers that are wrong. I build research and auditing software designed to catch that before you read it: several models answer independently, a synthesis step reconciles them, and it is allowed to conclude there is no consensus.

N° 01 — Ventures

A small portfolio of
ambitious things.

What I’m building, and where each one stands.

Opportunity HunterN° 01LIVE

Opportunity Hunter

Research engine that sources every claim from a live page.

Research
Coffee Be InN° 02CLIENT BUILD

Coffee Be In

PWA for a coffee shop in Patras.

Awaiting final content from the owner.

Web
ClinicOSN° 03IN DEVELOPMENT

ClinicOS

Practice administration software for private clinics in Cyprus and Greece.

Healthcare administration
MeltemiN° 04DEMO

Meltemi

Direct-enquiry app for villa stays.

Demo with sample villas and AI-generated images.

Web
DataExtractN° 05LIVE

DataExtract

Turns PDF invoices into structured data, ready to check and import.

Bookkeeping
MerakiN° 06DEMO

Meraki

Event-quoting app: guests build an event and see an estimate as they go.

Demo: a fictional events company with sample prices.

Web
N° 02 — Method

Built to prove
its own output.

I build with a system I wrote myself. It runs research, auditing and analysis — and its main job is establishing whether its own findings are actually true.

01

Verification before assertion

Every claim is checked against a live source before it is reported. A page counts as read only if the HTTP status is valid and the content is not a disguised error page.

02

Independent voices

Up to five models answer the same question without seeing each other. A synthesis step reconciles them, names the model behind every argument it cites, and is allowed to conclude there is no consensus.

03

Machine-readability auditing

A thirteen-point audit of how legible a site is to an AI assistant: structured data, semantic markup, canonical addressing, machine-readable summaries.

NEVER INFER
SUCCESS
NEVER INFER
COMPLETENESS
NEVER INFER
A CHANGE
VERIFY AGAINST
GROUND TRUTH
04

Four rules

A request that returns is not a request that succeeded. A file with the right headings can still be two-thirds missing. An edit that matches nothing exits cleanly. Every check is validated against a real page whose correct answer is known independently.

How it works
Why the studio exists

Most software fails long before the code does — in the decisions nobody wrote down. I build small, deliberate systems that hold those decisions, so the work keeps running when the room is empty.

Civil engineer. Twenty-four years co-owner and general director of a trading company in Greece. In Cyprus since 2012 — now building software in Nicosia.

Founder, North Star LabsNicosia, Cyprus
N° 03 — Contact

Get in touch.

If something here is useful to you, write to me directly.

contact@nstarlabs.com
CompanyNorth Star Labs Ltd
StatusUnder formation
Based inNicosia, Cyprus
Emailcontact@nstarlabs.com