Leuven, 03/06/2020 - The founders of the Belgian scale-up Dataroots specialized in AI and machine learning, launch the platform dploy.ai. Companies rarely achieve smooth integration of their AI models, despite their investments in artificial intelligence. Moreover, legislation and regulations require that predictions of these models are entirely auditable. dploy.ai can be implemented as a platform-as-a-service in no time to minimize the step between experimentation and ROI. It maps everything using an online dashboard and various modules such as performance monitoring. This way, the platform ensures fast integration for companies in a compliant, robust and scalable way. A free preview version is available to test the platform.
New guidelines insist on the auditability or verifiability of AI models in order to manage the rise of artificial intelligence and to protect the customer whose data is used. For example, when a bank employee performs a credit check using AI, his computer program consults an AI model which answers with an estimate. Through new guidelines, the potential customer can, to a certain extent, demand that this calculation is reviewed and made transparent.
However, meeting these requirements is a major challenge as there are little to no standards on how to communicate with AI models, how to monitor performance over time, how to perform an AI model audit or how to make a model explainable. "That's what dploy.ai is changing now," says Bart Smeets, CEO of dploy.ai. "The platform is unique of its kind because it makes a transparent connection between the AI model and the user. Everything becomes traceable: the predictions that were made, by whom, with which version of which AI model and to which level of performance. Compliance & risk departments are thus completely compliant in terms of auditability. Moreover, the model is deployed in a production environment that is robust, scalable and secure.”
Many companies that invest in artificial intelligence and the development of AI models find it difficult to integrate these models into their IT systems, resulting in a lack of return on investment. "dploy.ai largely unburdens companies by providing them with the infrastructure, follow-up and compliance data out-of-the-box. This allows them to focus on building models while benefiting from a direct operational profit and return on investment," says Bart Smeets.
The development of dploy.ai grew organically from the way Dataroots operates. "To ensure the quality of the AI model and a successful integration, we use a fixed methodology in every project. This also includes setting up the infrastructure and the connection between the model and the user, which until now has been in function of a specific project. With dploy.ai we have now bundled this into a platform-as-a-service,” says Jonas Tundo, CEO of Dataroots.
Free preview version
A first preview version was launched on the dploy.ai website. This version is free for individuals and companies to use and test a couple of pre-made models. In the coming weeks and months, dploy.ai will allow deploying proprietary models and monitoring them. Feedback from users will also be taken into account in order to respond to their questions and needs as much as possible.
There was also the launch of a first pilot project with an online retail company that carries out approximately 20 million predictions per day. Bart Smeets looks ahead optimistically: "We thoroughly believe in the power of dploy.ai. We are launching an end-to-end platform that brings fast, secure and regulatory compliant AI deployment to users’ doorstep. Soon, dploy.ai users will be able to unleash the capabilities of their AI models without the usual difficulties.”