Abstract gradient with rounded corners, blue and purple color scheme background.

Catch drift in robots and AI, the drift others miss

See robot results

Information monitors for robots, RL agents and LLM pipelines: wear and drift detected from the data your systems already produce. No training data, no model access, no retraining.

A research lab for complex-system monitoring

Semarx develops monitoring methods from information theory. Instead of modelling a system, we measure how tightly its inputs, actions and outcomes stay linked, and flag when that link weakens. The method has been tested on robots, RL agents and LLM pipelines, and each result is published as a preprint.

We don’t sell a finished product. We provide the method, architecture, algorithms and engineering capability to help teams build, tailor and scale information monitors within their own systems.

Drift: What Accuracy and Reward Do Not Show

Scores can look normal while a system drifts. The information monitor checks whether what the system sees, does and gets back still fit together, using only data the system already produces.

Comparison: usual monitoring checks each signal or score on its own; the information monitor checks whether what a system sees, does and gets back still fit together, using only normal operation data from the system itself, and indicates roughly where a change occurred.

Illustrative schematic: the task score stays flat while the system drifts; the information monitor departs from its healthy baseline when drift begins, well before the score drops.

Diagram: an AI agent's actions and observations. When it stops working, the information monitor helps tell whether the agent changed (actions side) or its environment changed (observations side).

ABOUT US

Measuring Drift with Information Theory

Information theory measures how much one signal tells you about another, with no model of the system needed. We use it to track how strongly inputs, actions and outcomes still inform one another, against the system’s own healthy baseline.

The Information Digital Twin (IDT)

Drift Detection for Machines and AI

Information Digital Twin architecture for non-invasive AI assurance.

The IDT is our drift detector for autonomous systems. It runs alongside your robot, RL agent or LLM pipeline, reads only what goes in and what comes out, and computes information-theoretic metrics from that stream.

When drift appears, it gives a first indication of where, on the sensing side or the acting side, so your team knows where to look.

At its core is the information monitor, available today for robots, RL agents and LLM pipelines: no model access, no retraining.

The Human Digital Twin (HDT)

Drift Detection for Human-Machine Teams

The HDT applies the same information monitoring to the interaction between a human operator and their system, in cockpits, control rooms or clinical settings.

It reads the interaction stream: actions, observations and outcomes. When operator and system stop predicting each other, the HDT flags the drift.

Systems can then simplify the interface or alert a supervisor while the situation can still be recovered.