Hummingbird AI
Helping enterprises and governments scale AI solutions: low emissions, low cost
Introduction
We revolutionize the future of AI by creating streamlined models that maintain accuracy with minimal losses while being 100X faster and more compact. This not only slashes computational costs but also ensures swift real-time decision-making in critical sectors like healthcare and finance. What the smartphone was for large computers, we are for AI.
Problem
Most AI/ML models are bulky and require significant computational resources. This limits their deployment in real-time applications and increases both operational costs and emissions. Most focus in the industry targets either (1) new use cases without regard to the limitations in the hardware or (2) improved chip design which neglects the software and focuses mostly on training costs and speed rather than inference.
Opportunity
Prune down AI/ML models to a fraction of their size while preserving 96-100% of their accuracy. We effectively create (1) AutoML APIs that get rid of wasted data within the neural network and (2) microchips with generalist models pre-loaded for off-the-shelf use (e.g. image recognition model on a small camera). This solution allows for quicker inferences, reduced storage needs, and the capability to run models offline in critical scenarios.