Decisions for
the physical world.

Action models that unlock optimal decision-making for complex real-world operations.

Get in touch How it works

Omenic (adjective)
Having the quality of an omen

Problem / The gap

Why plans break in
the real world.

Three forces work against every operational plan.

01 / 03

Uncertainty.

The future is a spread of possible states, not a single forecast.

02 / 03

Cascading effects over long horizons.

A small disturbance compounds into a system-wide disruption.

03 / 03

Uncontrollable external factors.

Forces outside your control can sweep the plan away.

Solution / Overview

Enabling operators to rely on intelligence, not instinct.

How? Through two key capabilities.

01 Simulate

The ability to efficiently simulate thousands of decision scenarios, despite uncertain physics.

02 Optimize

The ability to optimize over these scenarios to find the optimal decision trajectories.

Solution / How it works

World models, coupled with reinforcement learning.

We build operational world models: a foundational AI physics surrogate paired with an operations model trained on each site's data. Reinforcement learning then searches thousands of imagined rollouts for the decisions that hold up best, before anything happens in the real world.

Status Acting Scenarios 2,048 Policy iterations 512

Decision trajectory Illustrative

Operational World Model

Physics Backbone

Foundational AI surrogate model

Predicts how a physical system evolves, fast enough to explore thousands of possible futures.

Operations Model

Trained on operational data

Captures how a system responds to each action given the dynamics.

Reinforcement Learning

Policy optimisation

Finds the decision trajectories with the best outcomes across imagined rollouts.

Starting point / Weather

We start with weather.

The most consequential physical system on Earth, and the most observed. Two numbers explain why, and why now.

01 / The cost

Routine weather costs the EU economy

€400B¹

every year, much of it through suboptimal planning.

¹ Allianz Global Corporate & Specialty (2013).

02 / The unlock

AI weather models are up to

10,000×²

more efficient than traditional simulation tools, yet that efficiency has stayed trapped in research.

² Bi et al., Pangu-Weather, Nature (2023).

Acting is the new AI frontier.

No one owns the acting layer for weather-sensitive operations yet.

Domain	Wave 01 / Predict	Wave 02 / Act
Engineering AI	Established	Emerging
Computer vision	Established	Emerging
Weather	Established	Unclaimed

Landscape / Competition

We go beyond forecasting.

Forecasters tell you what the weather will do. Omenic tells operators what to do about it, pairing AI weather models with actual decision-making.

AI weather forecasting Traditional forecasting Indicators Actual decisions AI weather forecasting Weather intelligence platforms Weather data APIs National weather services

Traction / Pilots & grants

First pilot live, more pilots in the pipeline.

Aviation Live
Energy Live
Insurance Later
Logistics Later

Grants

CHF 190K

in non-dilutive funding from Innosuisse, Venture Kick, Talent Kick and BRIDGE.

Team

Born out of research at ETH Zürich.

Over ten years of combined research, peer-reviewed in top machine learning and engineering venues. The right expertise for both layers of the stack: the physics world model and the decision engine.

Layer 01 / World model

Gregory Duthé

Co-founder & CEO

PhD, ETH Zürich. Geometric deep learning for surrogate modelling of physical systems.
Layer 02 / Decision engine

Giacomo Arcieri

Co-founder & CTO

PhD, ETH Zürich. Reinforcement learning for decision-making under uncertainty. ETH Medal.
Open

Join us.

We are growing across ML research, forward-deployed engineering, infrastructure and commercial.
Get in touch