Selected work

Where it had
to be right.

Work where a mistake would cost money, time or health, and what we did to make AI dependable there. Honest accounts, no invented numbers.

Health research · Eurostars

SPEC-NEURO

Coordinating a Eurostars consortium that brings non-invasive monitoring to neurological care.

Measuring pressure inside the skull usually means surgery, and diagnosing a sleep disorder means a night in a lab. SPEC-NEURO develops a non-invasive alternative, with AI as the bridge between the sensor and a number a clinician can trust. Xantage coordinates the consortium, with Uppsala University, Yonsei University and Dankook University among the partners. The models are built to be explainable and are validated against invasive gold-standard measurements in hospitals in Sweden and Korea.

What we did

  • Consortium coordination
  • Explainable AI and clinical validation design
  • Privacy-by-design data architecture
Decide · Build · Accelerate

Alfa Laval

Three engagements over several years, each built on what the last one established.

Large organisations rarely lack data or opinions. What they lack is one version of each that everyone trusts. Our work with Alfa Laval has moved from data, to decision, to delivery.

What we did

  1. Build Self-service, on data you can trust

    Smaller machines brought many more customers, and not all of them buy on-site service. Alfa Laval wanted a self-service portal where customers find the right service kit for their machine. The hard part was inside: product data spread across many systems, with the same part identified differently in each. With IT and the business, we set a product data strategy. Then we built the pipeline that cleans, reconciles and describes that data, from source systems to the customer portal. It covers decades of service and repair data for major product lines. On that foundation, we prototyped an early AI assistant for one product line.

  2. Decide One picture of the problem

    Everyone agreed something was wrong. Few agreed on what. Each department saw part of the picture, coloured by its latest customer conversation or claim, so the proposed fixes pulled in opposite directions. We interviewed experts across headquarters, then the sales organisations closest to customers. The views turned out to be less contradictory than they sounded. Traced back, many symptoms led to a few root causes. That evidence became the shared basis for what to fix first.

  3. Build · Accelerate Closing the gaps

    Now we are building the fixes for the gaps the research found in processes, documentation and software. That includes a stronger software development lifecycle and AI-assisted coding for the development team. The wider brief is AI enablement: finding processes across the organisation where AI removes manual work and raises quality at the same time. Work gets faster too, but quality is the goal.

Water industry · Decide · Build · Run

Private AI cloud

A dozen AI ideas, narrowed to the ones that pay off, then built on a private AI cloud that we run.

One engagement covered every step, from the first list of ideas to a system we run and develop today.

What we did

  1. Decide Choosing where AI pays off

    The client had more ideas for AI than it could pursue. Together we assessed a dozen use cases, each for feasibility and business value, and agreed which ones to build first.

  2. Build A private AI cloud

    We designed and built a private AI cloud, so the chosen use cases run on the client’s own data without that data leaving its control.

  3. Run Run and developed by us

    At the client’s request, we run it: monitored, updated and dependable, with new capabilities as needs change. The client owns the system and can take it over at any time.

Accelerate · Spark

RLC Institute

Spark, an AI coaching platform built with the Rapid Learning Cycles Institute.

Rapid Learning Cycles help R&D teams make key decisions on evidence instead of on schedule. Spark puts the method into everyday work. It checks whether a hypothesis or a key question is framed well enough to learn from, helps write the reports the method depends on, and makes past learning searchable. It was built with the Institute, around how the method is actually taught and used.

What we did

  • Product built with the RLC Institute
  • AI evaluators for hypotheses and key decisions
  • Report writing and searchable archive
In good company

Organisations we have worked with.

Clients and partners
  • Alfa Laval
  • HOMA Pumps
  • Streamline Hospital Services
  • Uppsala University
  • RLC Institute
  • Yonsei University
  • Dankook University
  • NextAgents
  • Quantale Intelligence
  • Boroka

Working on something that has to be right?

Tell us where you are. We will be straight about how we can help.

Book a consultation