Alexandre Lebrun
CEO & Co-Founder
Built time-series learning systems for high-stakes physical environments. Led product direction from raw sensor modality to world-model architecture for equipment control.
Predict how your machine responds before you change the line.
AMI Labs learns a world model from your equipment's own sensor history, then lets you simulate any control change before an engineer commits it on the line.
Every setpoint change is a live experiment.
Process engineers on the line have no safe way to preview what happens when they adjust a control parameter. Every modification is committed directly to running equipment, with outcomes that only become visible minutes or hours later.
The cost is retuning cycles, production variance, and in the worst cases, damage and unplanned downtime. Expertise transfers poorly because the knowledge lives in individuals who have run the equipment long enough to build intuition.
How AMI Labs changes this
How It Works
Three steps from historian to prediction.
-
Connect your sensor history
Feed in existing data via OPC-UA connector or SCADA historian export. AMI Labs works from the data you already collect, no new instrumentation required.
-
Train the equipment world model
AMI Labs trains a causal model specific to your machine and its operating conditions. The model learns the relationships between control inputs and sensor responses for this equipment, not a generic industrial average.
-
Simulate before you commit
Query the prediction API or use the web interface. Ask what happens to vibration, temperature, and output quality if feed rate goes up by 4%. See the projected response trajectory before any engineer touches the line.
7/9
machine baselines in our early-access program: optimal setpoint found on first live attempt
2-5
trial iterations per setpoint, prior to using world model prediction
90 days
Pilot program duration with validated model and results report at close
Setpoint Simulation
World Model v2.4
Feed rate adjustment
+4.0%
Vibration (bearing 3)
+2.8%
at T+3.5 min
Pressure (inlet)
+1.2%
at T+1.8 min
Outlet temperature
-0.4 C
at T+7.2 min
Model confidence
93%
"Knowing how the reactor would respond before we touched the setpoint was something we had never had before."
Structured for pilot programs and full plant deployment.
Pilot
$4,900
/month
1 machine line, up to 10 sensors. 90-day program.
- 1 machine line
- World model training and validation
- Results report at 90 days
Most Popular
Plant
$16,900
/month
Up to 20 machine lines. Ongoing subscription.
- Up to 20 machine lines
- SCADA and historian integration
- Dedicated onboarding engineer
Enterprise
Custom
Multi-site, unlimited lines.
- Multi-site deployment
- On-premise model hosting option
- Custom SLAs
Built by practitioners.
The AMI Labs team has spent years building learning systems for physical environments where the stakes of a wrong prediction are real.
Sophie Marchetti
CTO & Co-Founder
Designed scalable inference pipelines over industrial sensor streams. Built the data pipeline and training infrastructure that powers the world model at production scale.
Theodore Baudin
Head of Research
Specialized in causal inference methods applied to equipment behavior modeling. Developed the core prediction engine that distinguishes AMI Labs from regression-based approaches.