Lab OS

Agents that understand the experiment.

We design autonomous loops for labs where the signal is faint, the instruments are many, and the human attention is expensive. Built for hydrogen R&D, fusion diagnostics, advanced reactors, and any experiment that runs in campaigns.

Architecture

A four-loop control system.

Every loop is instrument-agnostic. We connect to whatever you already have, wrap it in physics-aware interpretation, and expose decisions back to the hardware or to a human reviewer.

01 — Ingest

Multi-channel data ingestion

PMT, Timepix3, spectrometers, oscilloscopes, power supplies, PLCs, environmental sensors. We normalize streams, tag provenance, and flag missing or stale data before it reaches the interpretation layer.

02 — Interpret

Physics-aware anomaly detection

Agents model the expected instrument response from first principles. They know the difference between a meaningful regime transition and a power-supply glitch — because they know the physics of both.

03 — Decide

Policy-driven run control

The agent proposes the next action: continue, pause, sweep a parameter, repeat a run, or escalate. Policies are set by the principal investigator and can be tightened or loosened per experiment.

04 — Act

Closed-loop actuation

Setpoints are written back through safe APIs or SCADA gateways. Irreversible or high-consequence actions require human confirmation; routine optimization runs autonomously.

Safety layer

Hard physics gates before any claim.

Our Archimedes system rejects conclusions that violate dimensional consistency, conservation laws, or known experimental bounds. The agent does not hallucinate a discovery because the gate won't let it.

GATE A

Dimensional consistency

Every derived quantity is unit-checked before it is reported.

GATE B

Conservation residuals

Energy, mass, and charge residuals must close within defined tolerance.

GATE C

Existing-bound check

Predictions are compared against published limits before they become claims.

GATE D

Provenance log

Every decision, model version, and instrument state is archived.

Applications

Where Lab OS earns its keep.

Hydrogen & plasma electrolysis

Continuous monitoring of pulse parameters, gas output, and emission signatures — with automatic regime classification and next-run suggestions.

Fusion diagnostics

Multi-diagnostic correlation during shots: PMT, interferometry, magnetic probes, and bolometry — all checked against each other in real time.

Advanced reactor test loops

Thermal-hydraulic and materials experiments that run for days or weeks, with agents that pause on anomaly, not on a clock.

Materials synthesis campaigns

Closed-loop parameter exploration for catalysts, coatings, and intermetallics — combining in-situ measurements with Bayesian or physics-guided search.

Radiation & particle detection

Timepix3 and PMT networks with coincidence logic, noise rejection, and automatic calibration drift tracking.

Custom frontier setups

If your experiment has instruments, a hypothesis, and a person who wants to sleep — we can probably build a loop for it.

Why Nornec

We are not a generic AI consultancy.

Nornec is a physics-native practice. The agents we deploy are constrained by the same first-principles discipline we apply to hydrogen plants, nuclear systems, and frontier energy projects. We don't sell dashboards — we ship control loops that know when they don't know.

  • Principal-led, fixed-scope engagements
  • Code and runbooks you own
  • On-prem or private-cloud by default
  • No black-box subscriptions
Typical deliverables
  • 01 Instrument integration spec
  • 02 Physics-aware anomaly model
  • 03 Agent policy and escalation matrix
  • 04 Closed-loop deployment on your stack
  • 05 Runbook and hand-off training

Build the next version of your lab.

Start with a Lab Audit. In 48 hours we'll map where autonomy is ready, where it needs guardrails, and where humans must stay in charge.

Request a Lab Audit →