Luxxera

Context

Luxxera is a medical-tourism marketplace connecting international patients with vetted cosmetic clinics in Greece, covering discovery, consultation, booking, payments, and aftercare. It has three sides: a patient app, a clinic dashboard, and an internal admin layer for compliance and payment governance. It operates where GDPR, HIPAA-equivalent privacy rules, and Greek tax law meet.

I joined as co-founder and CTO when there was no team, no process, and no code. The company raised a $500K pre-seed and reached $100K MRR within six months of launch.

The team

The work expected of a company's first technical hire, done from a blank page.

• Built the hiring practice end to end and hired four engineers onto a globally distributed team: sourcing, screening, technical evaluation, and the final call.
• Designed an interview process a non-technical interviewer can run, so the bar no longer depended on me as its only judge.
• Owned onboarding, mentoring, and performance conversations as the senior technical authority. Escalations routed to me.
• Wrote the standards down as shared context rather than tribal knowledge: coding conventions, review norms, branching workflow, definition of done, and an ADR process. The same context guides the AI coding tools the team uses.

How we shipped

• PRD-first specifications, user-story decomposition and estimation, two-week sprints, design reviews, standups, and retros.
• A fully automated build, test, and deploy pipeline on GitHub Actions into AWS, which let the team ship features daily with weekly scheduled analytics rollouts.
• TDD and BDD on the platform, with AI used to validate and speed up test creation.
• An AI-driven intake system that classifies incoming product work and applies weighted priority, so planning starts from a sorted queue instead of a hand-triaged backlog.

Technical direction

I stayed in the code, and the calls I made were the ones a team can't easily reverse later.

• Guided the migration from a Next.js monolith to a NestJS API gateway in a Turborepo monorepo with hexagonal architecture. I prototyped the approach myself before the team committed, and the existing system kept shipping through the transition.
• Designed the PostgreSQL data model: multi-tenant schema, a documented patient-to-clinic booking state machine, and a dual-database split between application data and CMS content.
• Architected to GDPR and HIPAA-equivalent requirements on AWS: field-level encryption for PHI, least-privilege IAM with AWS Organizations and IAM Identity Center, WAF, a security review for new systems and vendors, and compliance tracked through Vanta.
• Published the field-level encryption package as open source (payload-plugin-encryption, AES-256-GCM) with key rotation and migration utilities.
• Shipped a self-hosted RAG system on AWS Bedrock and pgvector, chosen over managed vector services to keep cost, data, and compliance control over patient data. Negotiated a BAA with OpenAI so AI features could run inside the platform's PHI boundary.
• Stripe Connect payments with clinic payouts, refunds, and Greek e-invoicing compliance. Real-time clinic-to-patient messaging and video on Stream, modeled to scale past the vendor's team limit.
• Infrastructure as code with Terraform on ECS, S3, EventBridge, and Route 53, behind Cloudflare.

Outcomes

• $100K MRR within six months of launch on a $500K pre-seed.
• Six clinics onboarded out of more than fifty that applied. The low number was the point: a selective bar, not a ceiling.
• Consultation calendars booked out months ahead. One clinic left public-sector work to go full-time private on the back of Luxxera volume.
• A team that runs day-to-day delivery without me. I handed it off in July 2026, stayed on as an advisor through September, and left to find a larger engineering organization to lead.

What I took from it

• Write the standards down before the second hire. Context that lives in a document scales; context that lives in the founder's head does not.
• Prototype the risky migration yourself, then let the team own it. The cost of a week of prototyping is far lower than a wrong architectural bet.
• Owning the domain logic and treating third-party tools as execution surfaces made every vendor swap boring, which is what you want.