Jev and the Art of Making Up Your Mind

On the shuttle back from the opening reception at the Cross-VISS convening in Atlanta last week, a colleague asked me, “So what makes Jev special?” They pointed out that there are already a bunch of open-source models, including BERT, and finished with, “What makes this model worth my time?”

Fair question.

That conversation gave me a reason to spend more time with Jev. Calling it a decision model only gets me so far when someone is asking why they should use it. I need to understand what I can ask it to do, what comes back, and whether that answer helps with anything I actually care about.

I started with two things from my work: deciding where a research software issue belongs, and checking whether a source supports a sentence I’m writing. I made up the inputs for both examples so you can run the same requests and compare what you get.

TypeSafe calls Jev a decision model because you give it questions with defined kinds of answers. That might be a choice between issue labels, a yes or no, or a score on a scale you’ve described. It returns those answers with probabilities. From there, I can write ordinary code that decides which queue an issue goes into or whether a sentence needs another look. The API documentation shows how those questions fit into a request.