Evaluation runs across all four
Baselines, benchmark design, error analysis and human-review protocols. Every piece of work is signed off against a measure agreed before it starts.
A Swiss engineering company Over twenty years in business
Engineering software that has to keep working.
Artabel is a Swiss engineering company. We build information systems, industrial instrumentation and applied AI for industry, healthcare and cultural institutions. Our oldest system has been in daily production at an aluminium plant since 2008.
2004
Swiss Aktiengesellschaft, continuously on the commercial register in Canton Schwyz
4
Areas of work: applied AI, enterprise software, systems integration, instrumentation
17
Years the Confal enterprise platform has run in daily production
PIC •
Registered with the European Commission as a participant, PIC 868921733
Where our systems run
Aluminium smelting · SK
Semiconductor manufacturing
University hospital dermatology · CH
EPFL laboratory safety · CH
Real-estate asset management · CH
Technical university VR lab · SK
Open-air heritage museum · CH
What we do
Four areas of work.
A model that works on a laptop is not a system. Artabel covers the modelling and the engineering that puts it into service — integration, access control, release, and the years of maintenance afterwards.
PRACTICE 01
Applied AI
Retrieval and document intelligence, computer vision, speech and structured extraction, predictive modelling, data-centric AI.
- Source-grounded RAG
- Medical & industrial imaging
- Time series and forecasting
- Dataset quality auditing
PRACTICE 02
Enterprise software
Configurable workflow and information systems with role-based access, approvals, reporting and analytics, built to be maintained for a decade.
- Workflow & approvals
- Document circulation
- Reporting and dashboards
- Long-run maintenance
PRACTICE 03
Systems integration
Connectors, APIs, data pipelines, migration and deployment into estates that were never designed to be connected.
- API & connector engineering
- Release and change control
- Portability and export
- Security & access design
PRACTICE 04
Instrumentation & 3D
Sensing and smart devices in hostile environments, real-time tracking, plus 3D visualisation and immersive interfaces.
- Industrial IoT & robotics
- Real-time monitoring
- 3D and virtual reality
- Training environments
Programme delivery wraps them
Work-package leadership, deliverables on the month they are due, budget control, and the reporting and audit discipline European funders require.
Domain judgement on the project
Clinical governance for medical work, art history and digital humanities for heritage, and classical philology where sources are ancient. We name the specialists for an engagement before it is agreed.
Selected work
Three systems in service.
Client names, dates and what was delivered. The full list, including the engagements we cannot name, is on the work page.
Delivery footprint
Systems in production across Europe.
CH
Switzerland
Registered office, EPFL decision support, real-estate document AI
SK
Slovakia
Confal enterprise platform, Slovalco instrumentation and factory tracking, TU Zvolen immersive lab
INT
Beyond Europe
Semiconductor sector work
International project delivery
Engineering work packages in research consortia.
Artabel takes bounded engineering tasks in collaborative projects: interface specifications, a shared component built once and reused across pilot sites, integration and regression testing, and the deployment and cost documentation a review requires.
Shared engineering packages
One integration package, specified against the architecture the consortium agrees, tested for regression and fallback, then reused across pilot sites. Cheaper than rebuilding per partner, and comparable across cases.
Technical acceptance
Independent technical acceptance, security and accessibility verification, portability and export testing, a documented deployment guide and an honest maintenance and total-cost-of-ownership profile.
Programme administration
Separate task, time and cost records; declared conflicts and recusal; a single evidence register so no output or purchase is funded twice. Versioned releases and acceptance records, kept as the work proceeds.
Working in consortia
Engineering work packages in Horizon Europe consortia.
Artabel works in European collaborative research and is set up for it: registered with the European Commission as a participant, PIC 868921733, and organised to carry bounded engineering work packages in Horizon Europe consortia — interface specifications, a shared component built once and reused across pilot sites, integration and regression testing, and the deployment and cost documentation a review requires.
- Built to the consortium architecture — the scientific partner owns the schema and the acceptance criteria; we implement against them.
- Reuse over bespoke — one bounded package, many site configurations, with the maintenance cost stated.
- Negative results kept — stopped and failed cases stay in the evidence set.
- Human release gates — culturally or clinically sensitive material is published only on a named person’s approval.
Method
Evaluation is the part most projects skip.
Artabel’s practice is built on measurement: a baseline before the work starts, a benchmark that reflects the actual task, error analysis on the cases that matter, and a human review protocol for anything consequential.
The fields we work in — data-quality auditing, representation learning, benchmark construction, human–machine comparison under clinical conditions — are ones our director also publishes in academically. His university position is a separate appointment, described plainly on the research page; it is not a service Artabel sells.
Technical direction
Doctorate
PhD in artificial intelligence, University of Basel — deep learning in clinical dermatology, in collaboration with the University Hospital of Basel.
Prior degrees
MSc Computer Science, ETH Zürich. BSc Computer Science, EPFL.
Publication record
19 peer-reviewed publications, h-index 8, including work at NeurIPS and MICCAI and in the Journal of the European Academy of Dermatology and Venereology.
Funded research
Co-head of a Swiss National Science Foundation project on human and machine understanding of dermatologic conditions; lead of an Innosuisse project on synthetic-data benchmarking.
Earlier
Co-founder of a medical-device venture in India and of a data-science company in Dubai.
How an Artabel engagement runs
Scope, prototype, validate, implement.
Each stage has an exit condition and a decision point where stopping is a legitimate answer.
-
Scope — the question before the technology
We start from the task a person is actually doing, and how long it takes them today. What data exists, who owns it, what the rights and access constraints are, what a good answer looks like and who gets to say so. If AI is the wrong tool for the task, that conclusion is cheapest to reach at this stage.
-
Prototype — feasibility on your data
A working slice on your own material, sufficient to show failure modes, cost per task, latency, and how much correction work the system will create for your staff.
-
Validate — against the current method
The baseline is the existing manual process, the tool already in use, or a standard model. We measure task success, quality judged by qualified practitioners against agreed criteria, error frequency and severity, support burden and total cost. A faster workflow that lowers professional quality does not pass.
-
Implement — deployed, documented, handed over
Pilot deployment or integration into your stack, with role-based access, logging, approval gates, a manual fallback and a written deployment guide, so your own staff can operate and maintain it.
Sectors
Sectors we work in.
In each one, the professional judgement involved determines what the system is permitted to do.
Medical AI and MedTech
Imaging and multimodal interpretation, clinical decision support, structured documentation from consultation audio, research workflows and data-quality auditing.
Medical AI
Industrial AI
Lifetime prediction from fifty-plus sensor streams, predictive maintenance, anomaly detection, document intelligence and workflow automation — including molten-metal measurement above 700 °C.
Industrial AI
Agriculture and agri-food
Crop and quality monitoring, multimodal sensing, bioaerosol and particle classification, and and decision support for producers.
Agri-food AI
Culture, heritage and public knowledge
Source-grounded interpretation for museums and archives, rights-aware access tiers, provenance and attribution, multilingual access, and visitor-facing interpretation interfaces.
Heritage AI
Artabel also works on secure information systems, sensor analytics and critical-infrastructure support. Those engagements are not described publicly.
Next step
Tell us about the task.
A first conversation carries no charge or obligation. If you are assembling a consortium, we can provide a role description, an effort estimate and a partner profile within a week.