Jev powers streaming ad decisions with typed output under 200ms
Using early access to Jev, the author built an ad decisioning tool that feeds session data like current show, device and time of day, and outputs ad categories, audience segments and slot timings in well under 200 milliseconds.
I got early access to Jev, and to check its classification capabilities, I built a Ads decisioning tool that can be integrated with websites or streaming platforms like Netflix or Prime Video. I did the demo simulation of it on the basis of multi-axis session telemetry, feeding the model data like the current show, device form factor, time of day, and recent in-app browsing signals. I got results showing Jev evaluating all of these contexts in parallel to output perfectly typed ad categories, audience segments, and slot timings in well under 200 milliseconds. This makes it a much better fit for real-time ad-decisioning pipelines that have strict latency budgets, bridging the gap between rigid if/else rules and AI where traditional text-generating LLMs are simply too slow. However, the limitations include its heavy reliance on highly detailed schema descriptions to shape the logic, the strict legal requirement to implement user consent flows before gathering any real session context, and hard restrictions against inferring sensitive categories or running behavioral targeting on children's content. try-it-out:http://jev-demo-sandy.vercel.app
This post quotes another post: “ChatGPT co-inventor releases Jev, a model built for decisions rather than text”