Many AI features need to make a decision before they can do anything useful.
Should a request go to one workflow or another? Which category does this message belong to? Does an input need an extra safety check?
In an application, these decisions are easier to work with when the result has a predictable shape. A free-form paragraph can be useful, but application code often needs a typed answer it can route, display, or evaluate.
That’s the problem I’m exploring with Jev AI.
Jev AI lets you provide text or structured application state, define typed questions, and get structured results with probabilities. The goal is to make common decision tasks—like classification, routing, scoring, and safety checks—easier to connect to the rest of an application.
A simple workflow might look like this:
- Pass in a user message or relevant application state.
- Define the decision your application needs.
- Receive a typed result that your code can use.
For example, a support tool could classify an incoming request and use that result to choose the next workflow. A product team could score a response against a set of criteria before showing it to a user.
I’m building Jev AI as a decision model and API for software teams. There’s an online playground where you can try a workflow and inspect the request format:
I’d be interested to hear how other developers are handling structured decisions in AI-powered applications. Do you keep this logic in prompts, application code, or a separate service?
Top comments (0)