Models & Dark Factory01 Trust & Sandbox02 Research & Company03 News & Press04 Investors05 Job Board06

Built in the dark.
Built to be checked.

RMT Labs, a private research company in Luxembourg, builds its software in a dark factory — where AI agents code under human supervision, 45 people do the work of 2,880, and IM Catalina v2 ranks a million profiles in ten seconds.

Inspect the Models
DARK FACTORY · LIVE AGR-RED/25-017
64×
vs. Dec 2025
10 s
1M Ranked
24.6k
1i Titan Skills

AI agents coding · People supervising · Software shipping.

ARCHITECTURE & MODELS

Building is no longer the hard part.
Knowing what to build is.

Our factory builds the software; our people do the science. Output per person rose eightfold in January 2026 and eightfold again since — 64× compared with December 2025, allowing 45 people to operate at the scale of 2,880.

0×
Dark Factory Output

45 people doing the work of 2,880 vs. Dec 2025

0 s
1M Profiles Ranked

IM Catalina v2 across 119 match criteria

0%
Top-10 Precision

14% above a search-only baseline

0
Canonical Skills

71,678 aliases across 25 languages (17 Nov 2026)

0
Skillbourg Users

4,916 applications · 7,266 open roles (14 Nov 2026)

0
Peer-Reviewed Papers

PRICAI, ICAITD, EuroSPI & IEEE GLOBECOM

TWO MODELS CARRY THE WORK

Each part does one thing well.

Built to
be checked.

TRUST & GOVERNANCE

Tested in Sandkëscht, the AI sandbox of Luxembourg’s regulator.

From 31 March to 15 September 2026, Skillbourg participated in the CNPD AI sandbox (reference AIRS_2026-02). A sandbox gives guidance—not a certificate of conformity—and every compliance item below shows its real status today.

PROHIBITED PRACTICES
None identified

One-time assessment of one profile against one job posting. No permanent candidate score.

HUMAN OVERSIGHT
Zero automatic rejection

No threshold and no cut-off. Every applicant stays visible; a recruiter reads the explanation before deciding.

DATA PROTECTION
DPIA filed · 24 Sep 2026

Built with authority feedback and filed with the CNPD. Core platform data stored and processed in the EU.

FAIRNESS TESTING
Synthetic twin CVs

Names stripped before AI reading. Exit report highlighted synthetic twin-CV bias testing as a key good practice.

COMPLIANCE LEDGER · CURRENT STATUS & TARGETS
RESEARCH PROGRAMME

Four peer-reviewed papers.
One programme with LIST.

Approved on 9 October 2025 by Luxembourg’s Minister of the Economy, SME, Energy and Tourism (AgrRED/25-017, valid 1 Oct 2025 – 30 Sep 2028). We fine-tune our language models on MeluXina with LuxProvide and publish what we find—including the results that tell us to change course.

2024 – 2026 · ACTIVE

SafeHire

Ends Dec 2026. Matching architecture, experience normalisation, fairness & transparency frameworks.

2026 – 2028 · NEXT

TrustHire

Next phase extending verifiable fairness, multilingual compliance, and risk modelling.

2028 – 2030 · PLANNED

SmartHire

Long-horizon programme where a foundation model emerges as a by-product, not as the goal.

MINISTRY ACCREDITATION
Agrément organisme de recherche privé
CHRONICLE

From recruitment to research.
Built in Luxembourg since 2022.

A private research company of 45 people: 30 employees in Luxembourg and 15 contractors around the world — with new offices opening now in Saarbrücken (Germany) and Bucharest (Romania).

ECOSYSTEM

Part of Luxembourg’s
research & compute ecosystem.

RESEARCH & HPC

Science & Supercomputing

Luxembourg Institute of Science and Technology (LIST) and LuxProvide for MeluXina, Luxembourg’s national supercomputer.

INSTITUTIONAL SUPPORT

National Innovation

Luxinnovation, Luxembourg Chamber of Commerce, Luxembourg City Incubator, House of Entrepreneurship, House of Training, Luxembourg House of Cybersecurity & Digital Learning Hub.

TECHNOLOGY PROGRAMMES

Cloud & GPU Alliances

Microsoft, Google, Amazon Web Services, STACKIT, and Nebius supporting scalable European cloud and dedicated GPU workloads.

MEMBERSHIPS & HONOURS

European Network

Startup World Cup Luxembourg 2025 Top 10 Finalist. Member of FEDIL, Pulse, Paperjam Business Club, Silicon Luxembourg, IT Nation, AmCham Luxembourg, RomLux & bilateral chambers.

LEADERSHIP & BOARD

The people who lead it.

Governed by a three-member Board of Directors (Daniel Stoica, Raluca Stoica, and Maria João) with joint two-director signature on every commitment, supported by a six-member Advisory Board.

Daniel Stoica

Daniel Stoica

Chief Executive Officer

Founder, Board Director, and co-author on our EuroSPI 2026 unified risk paper with LIST. Leads company strategy, investor relations, and European AI vision.

Raluca Stoica

Raluca Stoica

Chief Partnerships Officer

Board Director bridging deep recruitment expertise with strategic partnerships across Luxembourg’s ecosystem, European institutions, and Skillbourg adoption.

Artem Antonenko

Artem Antonenko

Chief Technology Officer

18+ years architecting high-load distributed systems (50M+ daily users). Co-author of ResumeParsingBench (PRICAI 2026) and leader of our Dark Factory engineering organisation.

CHIEF FINANCIAL OFFICER · BOARD

Maria João

Board Director overseeing financial strategy, corporate governance, and scaling operations.

CHIEF OPERATIONS OFFICER

Mihai Popescu-Anghel

Leads operational delivery, multi-country expansion across Luxembourg, Saarbrücken, and Bucharest.

ADVISORY BOARD (6)
Stefan Tittel Melanie Babb Hans Troelstra Serge Borg Paul-Michael Schonenberg Bjorn Bronger
TWO DIRECTORS SIGN TOGETHER TO BIND THE COMPANY · RCS LUXEMBOURG B264777
NEWS

Lately at RMT Labs.

IN THE PRESS
Luxinnovation

Case study on RMT Labs and AI for recruitment.

AmCham Luxembourg

Interview with Daniel Stoica, CEO.

Startup Luxembourg

Taking a Luxembourg deep-tech startup international.

Chronicle.lu & IT Nation

Coverage of the TalentQuanta launch across EMEA.

Paperjam

Top 10 finalists of Startup World Cup Luxembourg 2025.

INVESTORS

The outline is public.
The detail is in the investor room.

RMT Labs builds ranking and matching models and sells them through its own products and services. Revenue of roughly EUR 2.1 million is expected in 2026 — more than double 2025.

Requests go to office@rmt-labs.com and are read by Daniel Stoica, CEO.

FOUR THINGS ANYONE CAN CHECK
  • Research: Approved by the Ministry of the Economy until 2028; 4 peer-reviewed papers with LIST.
  • Models: IM Catalina v2 (119 criteria) & 1i Titan (24,595 skills / 71,678 names).
  • Commercial: Skillbourg live since 6 Jul 2026; TalentQuanta placing ICT specialists; ~EUR 2.1M expected 2026 revenue.
  • Velocity: 45 people building at 64× output (equivalent to 2,880) in our human-supervised dark factory.
INSIDE THE INVESTOR ROOM
  • The current presentation
  • Financial statements and the forecast
  • Ownership and governance
  • Product roadmap
  • Go-to-market plan
  • Research and intellectual property
  • Legal and compliance documents
JOB BOARD

We hire with our own models.

114 open roles across RMT Labs, Skillbourg, and TalentQuanta (as of 14 November 2026). Every application is ranked by IM Catalina v2 with reasons, reviewed by a human recruiter, and answered with personal feedback.

01 · APPLY ONCE

Upload your CV. Skillbourg builds your profile from it, and you apply in one tap.

02 · RANKED WITH REASONS

IM Catalina v2 ranks your application for that specific role and writes down why.

03 · A PERSON DECIDES

A recruiter opens your profile and explanation before deciding. Zero automatic rejection.

04 · FEEDBACK EITHER WAY

You hear back with personal feedback, whatever the outcome.

Bring a ranking problem.
Your move.

We will show you what IM Catalina v2 and 1i Titan do with it — or open the investor room to inspect the research, financials, and governance firsthand.

FAQ

Straight answers, nothing hidden.

How our dark factory operates, how IM Catalina v2 ranks without automatic rejection, and how our research and CNPD sandbox work.

AI agents write our software under strict human supervision—up to 20 agents running in parallel—while our people decide what to build, conduct the science, and sign off every production release. By our own measure, output per person is 64× December 2025, so 45 people do the work of 2,880.

Never. There is no threshold, no cut-off, and no permanent candidate score. Every ranking starts from a single job posting, strips names before reading, explains every position in plain language, and leaves every decision to a human recruiter.

We tried fine-tuning one model to do everything, and it scored 50 out of 100. Separating canonical skill normalisation (1i Titan), machine-learning ranking across 119 criteria (IM Catalina v2), and language-model explanation raised top-10 precision to 86% while ranking a million profiles in ten seconds.

Core platform data is stored and processed in the EU, and our own language model is fine-tuned on MeluXina in Luxembourg. From 31 March to 15 September 2026 Skillbourg completed Sandkëscht (ref. AIRS_2026-02) with the CNPD: no prohibited practice was identified, our DPIA was filed on 24 September 2026, and synthetic twin-CV bias testing was cited as a key good practice.