AI systems — shipped & compliant Based in France

AI that ships.
Not demos.

I help companies across Europe turn “we should be doing something with AI” into systems that actually ship — and stay compliant while doing it. 20+ years of technical products across fintech, regulated data and blockchain.

EU AI Act ✓GDPR ✓Live products ✓
01 / AI Consultancy

Karven.ai

AI implementation across financial services, manufacturing, professional services, hospitality and real estate. Privacy-first pipelines: sensitive data is redacted on-premise before it ever reaches an LLM.

Deploy compliant AI →
02 / Programmatic SEO

Gadex.ai

End-to-end, AI-driven programmatic SEO. Multilingual content at scale — FR / IT / EN — to own the search results that matter. Without the thin-content trap.

Visibility that builds up →
Agentic AI Dev Blockchain SaaS Content strategy AI compliance

One standard

Useful

Systems that solve a real business problem and get used daily — not POCs that die on a shelf.

Defensible

Architecture and data that hold up in production, under load, over time. Fintech-grade rigor applied to AI.

Compliant

EU AI Act and GDPR built in from day one, not bolted on. On-premise redaction of sensitive data.

Background

Engagements carried out as a salaried consultant, before AlphaOscar. These organisations are not clients of Karven or Gadex.

Products

Products I design and run myself, grouped by theme.
See all products →

Who I am

Alexandre Orfèvre. Twenty years building technical products across fintech, regulated data and blockchain. Today I run two ventures — Karven.ai and Gadex.ai — with one through-line: AI and data products that are useful, defensible and compliant.

Frequently asked questions

Regulatory timeline current as of 10 July 2026.

Does the EU AI Act apply to my company?

If you develop, deploy or distribute an AI system in the European Union, yes. The timeline then depends on your use case: prohibited practices have applied since 2 February 2025, and general-purpose AI (GPAI) model obligations since 2 August 2025. The simplification package known as the “digital omnibus”, endorsed by Parliament on 16 June 2026 and given final approval by the Council on 29 June 2026, defers Annex III high-risk obligations to 2 December 2027 and Annex I obligations to 2 August 2028. Most enterprise use cases are not high-risk — but that has to be established and documented.

Can we send customer data to ChatGPT or another external LLM?

Not as things stand. Sending personal data to an LLM hosted outside your perimeter makes you a controller under the GDPR: you need a lawful basis, a defined purpose, a retention period, and — if the provider sits outside the EU — a transfer mechanism. The approach I deploy at Karven.ai inverts the problem: sensitive data is anonymised or pseudonymised on-premise before it leaves your infrastructure. The model never sees the identifier.

How long from POC to a system actually in production?

Six to twelve weeks for a first system in daily use, provided the data exists and is reachable. What takes the time is almost never the model: it is data access, permissions, integration with existing systems, and handling the edge cases.

Do you deploy models on-premise?

Yes, and it is the core of the offer. Open-weight models running on your own infrastructure or in an EU cloud, redaction applied upstream, and no sensitive data handed to a third-party provider. It is often the only arrangement that makes AI on regulated data defensible.

Who is behind AlphaOscar?

Alexandre Orfèvre — twenty years of technical products across fintech, regulated data and blockchain, including BNP Paribas and Thales, and today CPO of Rekord AG. He runs Karven.ai (AI consulting and implementation) and Gadex.ai (programmatic SEO). AlphaOscar is a French SARL registered with the Paris commercial court under SIREN 910 288 299.

Let’s find out where AI actually earns its place.

contact@alphaoscar.fr — replies within 48 hours

Tell me about your project

Your details are used only to answer your enquiry. No sharing, no ad tracking.