EX AI HUB · KAROL SZÉKELY

I teach machines to read.

You get time to think.

Multi-agent systems · Sovereign EU stack · Open-weight architecture

Let's talk
SCROLL
WHAT I DO
01

Decision Intelligence

Multi-agent systems that read thousands of pages and deliver decision-grade synthesis in hours. For due diligence, M&A, and pre-investment screening.

+
02

Sovereign LLM Stack

Open-weight models (Mistral, EuroLLM) in EU jurisdiction. On-prem or SecNumCloud. No vendor lock-in, no data leaving the EU.

+
03

RegTech & EU AI Act

FRIA scaffolds per Art. 27, DORA register-as-a-service, GDPR DPIA for AI systems. Ready for the August 2, 2026 high-risk deadline.

+
04

Entity Resolution

Splink + Neo4j for beneficial ownership graphs, sanctions and AML/PEP screening. Compliant with the 2024 EU AML package.

+
05

Multi-Agent Orchestration

LangGraph with validated message passing. Agents working autonomously 24 hours, delivering results before you turn on your computer.

+
06

Strategic Advisory

Partner-grade advisory for boards and investment committees. AI architecture review, risk assessment, scope alignment before PoC.

+
"AI isn't here to replace judgment. It's here to make judgment possible — for decisions one person alone can't read through."

I build multi-agent systems that read documents, connect entities, and deliver decision-grade synthesis to people in hours, not weeks. The stack is open-weight, in EU jurisdiction — no vendor lock-in, no data leaving the EU. I work best as a team of one: more code on weekends than PowerPoint on weekdays.

TRANSITION
Weeks of reading DD materials → Hours of decisions with an audit trail
AI vendor lock-in via US APIs → Sovereign open-weight stack in the EU
Compliance as reactive firefighting → EU AI Act and DORA as infrastructure
Let's start
MILESTONES
1998 First computer.
2008 First management roles.
2014 Independent advisory practice.
2022 First multi-agent systems.
2026 EX AI HUB · Decision Intelligence.
PRINCIPLES
01

Sovereign by default

Open-weight models in EU jurisdiction. On-prem or SecNumCloud. No vendor lock-in.

02

Decision-grade

Every claim falsifiable. Auditable decisions, citable sources.

03

Multi-agent over monolith

Parallelism, specialization, observability. No single-LLM black box.

04

Compliance from day one

EU AI Act, DORA, GDPR as infrastructure, not afterthought.

CONTACT

Let's do it

I'll respond within 24 hours

This site is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.