The AI Engineer's Guide to Surviving the EU AI Act

Appendix A. Designing AI-Powered Applications

This appendix explores a general approach to designing AI-powered applications—software applications that incorporate machine learning models to power one or more of their core features.

In recent years, there has been growing interest in the concept of “AI products” and the application of “product thinking” to ML- and AI-enabled systems. This approach focuses on AI capabilities not just as technical components, but as integral features that solve real user problems. Figure A-1 shows an example of an AI product: the O’Reilly Learning Platform’s generative AI assistant, Answers. This assistant is designed to help users interact with book content and conduct research across the publisher’s book corpus using natural language queries.

I define an AI product as a software solution that:

Delivers value

AI products provide meaningful value to users, stakeholders, or systems, whether by automating tasks, enabling new capabilities, improving efficiency, reducing costs, enhancing user experience and interactivity, boosting productivity, or generating new content.

Incorporates AI technologies

AI products utilize technologies such as machine learning, deep learning, natural language processing (NLP), and computer vision to enable or enhance functionality.

Interacts and adapts

AI systems can interpret and respond to inputs (such as data, user interactions, or other systems) in a way that is perceived as intelligent, and often they can learn ...