Edge AI Lifecycle

Teraki’s AI powered pre-processing improves the accuracy of sub-sequent AI-models by 10-30%. Achieved by efficient edge processing for the extraction of high-quality data (without noise) in normal and difficult conditions.

Teraki’s algorithm reduces latency by 10x for detection algorithms, typically 10-30 ms latency on ASIC. This enables real-time operations.

Teraki’s pre-processing consumes minimal power and reduces CPU footprint. The energy consumption for AI model is reduced with factor 10x. Efficient power consumption is essential for longer mileage range of the car/drone/robot.

Teraki’s pre-processing software is designed to operate in embedded devices and SoCs. It can run on various operating systems and is provided as a code-wrapper. The software achieves 4x -10x less data needed to train algorithms; hereby creating a suitable environment for complex AI-models to be operated on embedded hardware.

Teraki supports a continuous learning loop for customers to train and update pre-processing and their AI-models at the edge. The Platform enables customers to (re)train their own ROI/TOI models; refine the next data collection; and subsequently deploy more accurate AI-models. This process is automated to rapidly train new AI-models as well as to support AI-models that run in high-scale production.