POAM Foundation Models
Pre-trained Online Adaptive Models designed for sequential systems with sparse feedback, heterogeneous metrics, and regime-dependent dynamics.
What is POAM?
POAM separates system behavior into two parts: a shared latent state that captures underlying dynamics, and variable-dimensional observations that depend on the domain. This enables adaptation to new regimes without retraining, fast personalization, and robustness under sparse data.
Online Adaptation
Maintains belief over latent state for real-time updates
Regime Awareness
Implicitly represents regimes through latent dynamics
Variable Dimensions
Handles different metric dimensionality per system
Sparse Data Robust
Works with incomplete and delayed observations
Model Variants
POAM is a family of models with different capacity and deployment targets
POAM-Nano
Latent Dimension: 16-64
Optimized for edge devices and embedded systems. Ideal for small batteries, IoT sensors, and low-power applications.
Edge deployment
Inference-only mode
Low power consumption
POAM-Small / Base
Latent Dimension: 64-256
Designed for industrial systems and cloud inference. Supports active decisions and maintenance planning.
Industrial systems
Cloud inference
Active planning support
POAM-Large
Latent Dimension: 256+
Built for simulation, research, and pretraining. Used to learn transferable degradation priors across domains.
Simulation environment
Research applications
Transferable priors
Full Stack Data Infrastructure
Beyond our foundation models, we provide a complete data infrastructure for deployment, including simulation-driven pretraining pipelines, real-time data collection, and benchmark evaluation systems.
Digital Simulations
Generate diverse regimes at scale with controlled degradation mechanisms and realistic noise injection.
Data Collection
Long temporal coverage with multiple operating regimes and heterogeneous metric dimensionality.
Open Benchmarks
Evaluate adaptation speed, stability under missing data, uncertainty calibration, and robustness to noise.
Predictive Insurance Coverage
Through our insurance partners, predictive maintenance coverage is available to back our predictions. If our models fail to detect a failure within the specified window, you may be eligible for coverage of associated costs.
Insurance coverage is provided by third-party insurance partners and subject to their terms and conditions. Vertoaeris does not provide insurance directly.