Modern software is usually not build from zero - to integrate complex functionality, like Artificial Intelligence (AI), payments or text message notifications, today's software development teams utilise "building blocks", so called APIs that let them utilise this functionality build by a third party. The third party then also takes care of all the complexity of running and maintaining those specialised functions.
Computational Magic is a collection of APIs to allow for easy and fast integration of AI functionality, such as text summarisation, speech recognition or to extract text from images. We build it in a way, that allows your team to utilise our AI models without the need to supply us with any data, something other services require for fine tuning.
Furthermore, your teams can use any programming language and just a few lines of code, which means they can deliver AI enabled apps very quickly, saving you many months of work and costly resources.

You want to add AI functionality, such as text summarisation, speech recognition or image-text-extraction to your app? We can provide the right model for the task and take care of maintenance and operations for you. Computational Magic gives you an easy to use GraphQL API that allows you to interact with our AI models and utilise their power in your software just as easily as you'd communicate with your back-end. This also means you're free to use any programming language and framework without constraints, as all you need is web requests. Furthermore, our models are build in a way that doesn't require any training data from you for fine tuning, further lowering your barrier to get going and experiment with Computational Magic.

Transformer models are a big revolution for deep learning. We are building a suite of transformer models, fine tuned for different tasks, but optimised to be generalisable across multiple domains within that task. Paired with a GraphQL API, this allows for quick and easy integration in any project, regardless of programming language and without the requirement for training data for fine tuning.
Furthermore, we take pride in building our models in-house, integrating bleeding edge research findings and solving some computer science challenges we encounter along the way. This approach allow us to be market leading in minimising bias and maximising cross domain performance. The basis for our work right now are three core models: one language model, one speech-to-text model and one image classification model. Based on these, we create different task specific, fine tuned versions, using some special, internally developed methods.