Bigger inputs.
Smarter models.
Smaller footprints.
Embedded software that breaks through constraints in CPU/GPU processing and memory usage, training high data-intensity models on memory limited GPUs (requiring hundreds of gigabytes) and inference on edge devices in single-digit megabytes (requiring hundreds of megabytes).
01
Edge ML in Constrained Devices
Multiple CNNs running in parallel at the edge with minimal CPU/GPU and memory usage.
02
Compression of time series data
Exceptionally high compression ratios with minimal loss of fidelity makes transmission and storage of data more efficient and less expensive. Great for real-time digital twinning.
03
ML Training & Inference for Extremely High Data Intensity
Up to 4K pixel arrays run on standard object detection models; more data means better learning and higher accuracy.
Real World Applications
Utilities:
Grid-Edge Intelligence
Real-time anomaly detection and classification for predictive grid maintenance. High-fidelity compression for aggregation and
transmission of comprehensive data sets.
Industrial:
Electrical Panel and Motor Current Analysis
Condition-based monitoring of on-premise electrical infrastructure and powered equipment – via detection of anomalies and real-time digital twinning.
Health and Wellness
Enabling edge AI for resource-constrained devices; enhanced ML and inference for extremely high data-intensity applications.
At B Acceleration Week 2026 in Bilbao, Gordian gave live demonstrations of its compression and inference engines. The compression engine shrinks the time-series data footprint on edge devices, while the inference engine delivers low-latency machine learning directly on the device, making data-intensive applications more efficient and scalable.
The team also met one-on-one with corporates, technology partners, and investors to explore use cases in predictive maintenance, digital twinning, and industrial automation.
Event Spotlight
Gordian at
B Acceleration Week 2026
Why Gordian?
Edge ML inference
with minimal CPU/GPU and memory usage and power draw.
Shared memory across multiple models
for extremely high data-intensity model training.
Workflow friendly
drop-in embedded software layer, no model customization required.