I build products where machine learning meets everyday problems — from shared living in Greek cities to bystander CPR when every minute counts. These are active ventures I co-founded or lead as technical co-builder, not side demos.
Active ventures
Roomatos
AI-first roommate matching for Greece
Roomatos helps people find compatible roommates by lifestyle — not just rent and postcode. I co-founded the product and build the compatibility scoring and matching engine that powers Explore, Roomatch, and group household flows.
What we shipped
- Weighted matching — location, budget, habits, and lifestyle signals with explainable pairwise scores for 1:1 and group applications
- Full product surface — explore & filters, inbox, trainings with XP/badges, house groups, rent tracker, digital agreements
- Monetization — Stripe premium, freemium Roomatch daily limits, bilingual Greek/English UX
- Production platform — Next.js 15, Supabase (32+ tables, RLS), Vercel; GDPR audit and Playwright E2E
Stack Next.js · TypeScript · Supabase · Stripe · next-intl · TensorFlow.js (verification)
hygeIA
The Uber of First Aid — workplace SOS · AI · compliance
hygeIA is an end-to-end workplace emergency platform for Greece — one SOS tap logs GPS, alerts the floor, occupational physician, and manager in under 10 seconds, then opens real-time Greek voice guidance through ERC 2025 protocols. Medical steps come only from a hardcoded protocol engine; Meltemi classifies voice intent and never generates clinical advice. Works offline in warehouses, basements, and at sea.
Co-built with Papageorgiou Vasiliki (MD) and Skreka Theodora (Occupational Health). Pitched at Archimedes Accelerator Demo Day (Athens, June 2026). Evolved from EIT Health i-Days 2025 (2nd place nationwide).
What we built
- SOS → response pipeline — one-tap emergency, push/SMS alerts, AI guide live in ~10s, admin EKAB dispatch with one tap, automatic accident log book
- Greek AI assistant — Meltemi NLU with offline edge path; two-way voice so hands stay on the patient, not the screen
- ERC 2025 real-time guidance — JSON state machine, voice confirmation, injury photo upload with AI triage assist
- Compliance by design — Law 5239/2025 alignment, HSE admin dashboard, certification expiry alerts, PDF reports for inspections
- High-risk verticals — factories >250 staff, Piraeus shipping fleets (STCW + offline), private schools
Stack React Native · Expo · Meltemi 7B · Supabase · Zustand · ERC JSON state machine · expo-speech
Archimedes Demo Day 2026 · Athens
How I work on ventures
Problem first
Greek market constraints — language, regulation, trust, and infrastructure — shape every technical bet.
Safe AI boundaries
LLMs for understanding messy input; deterministic systems for anything high-stakes (matching rules, ERC protocols).
Ship to learn
Production deployments, real users, and measurable outcomes — not slide-deck prototypes.