App-First AI

In the application of ML in industry, one often faces the question, “I have a bunch of data. Can we build a useful model with it?”, and only after doing so, is it asked whether that model can be used for an industrial application. While useful for strategic brainstorming, this is now practiced to an unhealthy degree. We have observed this phenomenon and call it the “model-first” mentality, and often this leads to applications that cannot fully meet industrial needs – or require a lot of extra effort to do so. Therefore, a better approach is needed to provide the anchor, framework, and modern tooling around the realization that fundamentally, we still need to start with an “application-first” (or “system-first”) mentality. That is, the AI system development process needs to encourage thinking of the application structure, flow, and utlity, and then, let that guide the development and function of models.

App-First AI is aimed at:

• creating a powerful yet easy-to-use framework for building and running AI applications.

• providing end-to-end workflow, from development to deployment to execution and feedback.

• ensuring ensure that the development process starts with human needs, business requirements, use-cases and user-experience design, followed by modeling and construction.

• choosing and applying engineering best practices in building applications.