Applied artificial intelligence research

We move advances in artificial intelligence into working companies.

Most of what is published about AI in business is written by whoever sells the tool. This laboratory studies the same question from the other side: which techniques hold up when they meet a real process, real data and a person accountable for the outcome.

Field
Applied AI, decision systems, multi-agent architecture
Method
Study, prototype, production, measurement, publication
Scope
Five domain laboratories
Founded
2026
01 / Problem

The gap is not knowledge. It is transfer.

Statement

The distance between what artificial intelligence can demonstrably do and what a given company actually does with it is rarely explained by lack of technology. The technique is published, documented and reachable. What is missing is the work of moving it: understanding a specific process well enough to know which part of it a model can carry, and which part it must not.

That work is neither research nor consultancy. Research stops at the paper. Consultancy delivers a report and leaves. Both produce something that is read once. Transfer means the capability stays inside the organisation after we leave.

A laboratory is the structure that fits that job, because it lets the same finding be tested, corrected by production and published. A service company has no reason to publish what did not work. We do.

02 / Method

A closed cycle, not a delivery.

  1. Step 1

    Study

    A problem is admitted into a laboratory when it has appeared in at least three independent settings. Before writing code we establish what is already known, what the technique can support and where its hard limit lies.

  2. Step 2

    Prototype

    The smallest version that can test the hypothesis against real data. If it fails here it is discarded, which is the cheapest place to discard anything.

  3. Step 3

    Production

    What survives enters daily operation with named owners and defined measurement. This is where the failure modes no prototype reveals appear.

  4. Step 4

    Evidence

    Real use returns data: where it is right, where it is wrong, what edge case nobody anticipated and what it truly costs to run.

  5. Step 5

    Publication

    Whatever generalises is written up, always anonymised, including negative results. Submitting our judgement to someone who is not paying us is the only external check we have.

Fig. 1 — Transfer cycle. Steps 3 and 4 are what separate this from consultancy: the finding is corrected by the operation, not by the next meeting.
03 / Structure

Five laboratories, one per domain.

Domains fail differently. What you learn cataloguing spare parts does not transfer to formulating a cream, and neither resembles allocating an advertising budget. Each laboratory states its own philosophy first — what it studies and why — and then the projects that came out of it.

  1. LAB.01 Industrial Technical knowledge locked in catalogues and documentation
  2. LAB.02 Health & Pharma Assisted formulation and regulated decision support
  3. LAB.03 Marketing Reconciling campaign and revenue data for prediction
  4. LAB.04 Electoral Voting behaviour, party systems and forecast models
  5. LAB.05 Agentic Systems How multi-agent systems are designed to survive production
04 / Working together

How the laboratory reaches a company.

Routes

Three, and they are sequential: a resident programme that transfers judgement to a management team, a diagnostic that establishes which problem is worth solving first, and the development of the systems that come out of it.

None is sold as a package. Which one applies depends on what the organisation already knows, and that is what the first conversation is for.

Read the transfer model →