Nordic Edge AI Lab
Build Edge AI
Models for Any
Nordic Device
Custom Neuton models for CPU-run edge AI on any Nordic SoC and LiteRT models for Axon NPU
- Tiny self-growing Neuton models
- No-data wake word models
- No-code LiteRT model builder
Build models automatically using the Neuton neural network framework or configure your models using LiteRT.
Build in 3 Simple Steps
- 1
Upload Data
Upload training data or choose a wake word - 2
Train Model
Train automatically or configure architecture - 3
Deploy
Download model and run inference on device
Compatible with any Nordic SoC-based device
Build Intelligent Applications
Build AI solutions to recognize gestures and activities, detect anomalies, monitor vital signs, track and monitor assets, enable human-machine interaction in smart home and industrial systems.
Gesture
Wake word
detection
Anomaly detection
Human activity
recognition
Asset tracking and
monitoring
Health
monitoring
Smart Home
Interaction
Build Edge AI the Nordic Way
Build ultra-low-power, on-device AI optimized for Nordic hardware.
Optimized for Nordic
Low-Power Hardware
Designed for ultra-low-power edge devices
- Optimized for always-on sensing
- Balanced memory & performance
- Extended battery life
One Platform for
CPU & NPU
Build and deploy across hardware targets
- Neuton models to run on the CPU of any Nordic SoC (time-series & sensor data)
- LiteRT models to run on SoC featuring an Axon NPU (audio & advanced AI)
- Optimized pipelines for each target platform
Edge AI without
Complexity
No ML expertise required
- Automated model creation
- Fully on-device inference
- No raw data sent to the cloud
Prepare Your Data for Edge AI Models
Transform raw sensor streams into high-quality training data using built-in preprocessing tools designed for edge AI.
Windowing
Split continuous sensor data into time windows to capture meaningful patterns
Feature Extraction
Automatically generate signal features that help the model learn more efficiently
Feature Selection
Automatically identify and retain only the most relevant features for the model
Smaller models. Faster inference. Better accuracy.
Analyze Your Model and Data
Understand how your data & model performs and identify opportunities for improvement after training.
Model Quality Diagram
Visualize model performance across multiple metrics to quickly assess overall model quality
Feature Importance Matrix
Understand which features contribute most to model predictions and identify redundant inputs
Confusion Matrix
Analyze prediction accuracy across classes and identify common misclassifications
Data Analysis
Identify the most important features in your dataset based on statistical significance
Deploy Wake Word & KWSon Low-Power Devices
Create production-ready wake word & keyword spotting models in just a few steps
- No data
- No coding
- Ready-to-use on device
Build in 4 Simple Steps
1
Enter Wake Word
Ask phrase that activates your device
2
Train Model
Automatic training (~1 hour)
3
Test Model
Test the model directly in your browser
4
Run on Device
Download model and run inference on device
Bring Intelligence to Your Device
Build and deploy edge AI solutions with a platform designed for performance, efficiency, and long-term scalability