Stage 3 · Automate

AICertus seal

AI supports the process. The decision stays human.

AICertus™ is the method with which DaSP designs, governs and verifies AI agents in accredited laboratories: the laboratory’s knowledge, the expert’s verification, every action traced.

See the demoWhat the assessor sees
Knowledge Base
The laboratory’s knowledge, versioned: standards, procedures, more than twenty years’ experience in ISO/IEC 17025 and ISO 15189 laboratories, never generic data.
Human in the Loop
Every decision goes through a person.
Data stays in the laboratory
It does not feed public models: the installation belongs to the laboratory.
I tre pilastri: tecnologia AICertus, Knowledge Base, Human in the LoopAICertus™Tecnologiail metodo che governa gli agenti01Knowledge Basela conoscenza del laboratorio, versionata02Human in the Loopogni decisione passa da una personai dati restano nel laboratorio
Diagram of the method

Human in the Loop, in practice.

Seven practices, one flow. Choose a practice: the diagram shows where it lives.

  1. 01Documentstandard, SOP, test report
  2. 02Agentreads, compares, proposes
  3. 03Verificationthe expert checks
  4. 04Signaturethe decision stays human

if something does not add up, the agent tries again with the expert’s instructions

Data annotation

The laboratory’s people annotate the data the agent will use: what counts, what does not, in which version.

Signature responsibilities are not touched.

The life cycle of an agent.

Five steps that loop back on themselves, with human verification at the fourth.

  1. 01

    Risk assessment

    What can go wrong, with what effect on the quality system, before writing a single line.

  2. 02

    Controlled release

    On one process, with defined scope and thresholds.

  3. 03

    Continuous monitoring

    Every action recorded and compared with what was expected.

  4. 04

    Human checkpoints

    Where the expert checks, corrects, signs.

  5. 05

    Iterative improvement

    The expert’s instructions feed back into the agent’s program.

Custom Agent Blueprint

Every agent is born from a written specification: what it may do, what it may not do and who checks it.

What the assessor sees

A register of every action of the agent, who verified it and when; the documents the agent read and their version; the thresholds and the exceptions; the human decisions with name and date.

  • Data integrity according to the ALCOA+ principles

Material for the audit, not a risk to hide.

Agent actionDocument read and versionThresholdVerified byHuman decision
Comparison between standard and SOPStandard and SOP, version in forceAbove thresholdPerson who verifiesProposal accepted
Proposed updateDraft SOP, new revisionAbove thresholdPerson who verifiesAmended and approved
Case outside the patternRelated recordBelow threshold: stoppedPerson who decidesHandled manually
Final documentApproved version—Person who signsHuman signature
Diagram of the register, example entries: no names, no dates.

Designed in alignment with ISO/IEC 42001, ISO/IEC 27001, ISO/IEC 27017 and ISO/IEC 20000-1, the EU AI Act and the NIST AI RMF. Data processing compliant with the GDPR. The laboratory’s data does not feed public models.

The questions we get asked

Is AICertus™ a product to buy?

No. It is not a product: it is the method with which DaSP designs, governs and verifies the agents. You find it at work in the LABNEXUS agents, also when they work inside ATLAS LIMS.

Who is accountable if an agent makes a mistake?

Whoever verifies and signs, as today. The agent prepares, the person decides.

Does our data train a model?

No. No training of public models: the data stays within the laboratory’s perimeter.

See it at work

The method shows in the LABNEXUS agents at work: in the demo on our environment, in the pilot on one of your processes, with your records and the person who signs.

See the demoAsk for a pilot on your data

The demo: 30 minutes, on our environment, free. The pilot is a quoted activity: you start from a single process.

Back to the stage Automate