Ray

Keeping track of publicly available technology is an important part of honest machine learning consulting. Almost always, the best solution for a client is not for the consultant to build from scratch. In fact, an informed client should be suspicious of such a recommendation. Primarily, the consultant’s role is to build software connecting powerful tools to the specific operations of the client¹.

So we pay close attention to the AI/ML landscape.

Ray

A growing tool is Ray, the AI computing framework. It’s finding its place among leading companies in AI/ML applications:

• Uber used it for strong market optimization improvements and a boost to developer productivity.
• Instacart used it to improve resource utilization, developer productivity and reduced execution time.
• Pinterest used it to see similar gains.

This inspired us to get familiar with RAY, survey the positive and negative² use cases, and clean up our understanding into a shareable one pager:

Download PDF

---

Footnotes

1. In fact, this makes for no shortage of work. We often observe clients’ asks to be so specialized, no publicly available, designed-for-growth product quite does the trick. ↩

2. Few publicly express what failed, so we rely on our own experience, word of mouth and public discussions (X, reddit and GitHub issues mostly). ↩