AI toolchain SaaS solution to improve machine learning precision and reduce time and cost.

Offentlig prosjektsammendrag

Over the past 20-30 years, software developers have grown accustomed to having mature developer tools available to them. These tools cover all aspects of software development, e.g., early stage design, development and debugging, testing, release, and deployment, whether in the cloud or on a laptop. Contrastingly, the AI toolchain is hugely immature, and developers and data scientists have come to rely on a hodge-podge of poorly integrated tools and much manual guesswork. Data scientists, almost regardless of their field of work, struggle with many common challenges. Data scientists must acquire and identify the most appropriate data to learn from. Often thousands of hours are spent on feature engineering, which refers to loading, cleaning, contextualizing, discarding, and generally preparing data to be useable and useful for further machine learning. Furthermore, this data, consisting of tens of thousands of images, videos, documents, or combinations thereof, generally has to be labeled and appropriately categorized. In most cases, data scientists attempt to identify initial (pre-trained) models to start from, reducing the learning process considerably. Once the learning process starts, data scientists repeatedly have to monitor the process and attempt to optimize the learning process to arrive at the most appropriate model to use in production. Once the models have been made available and validated in production, data scientists will continue to be involved to understand critical issues such as data drift and model misbehaviour, which more often than not requires the model to be further (re-) trained involving a similarly long and tedious process to prepare new data or correct old data. Our solution aim to address these inefficiencies and help data scientists achieve their objectives in far shorter time, and with a significant reduction of cost.