Demystifying Decision Models: OpenJev, Decision 2.0, and Serverless Deployment on Runpod

​Not every AI task requires a model that writes paragraphs. Decision models skip text generation and return a probability distribution over the options you provide, answering in milliseconds with no malformed output and at a fraction of the cost of a chat model. That makes them ideal for routing, classification, guardrails, and the many small judgments agents make before acting.

​

​This hands-on workshop explains what decision models are, how they differ from LLMs, and when to use them, using two open implementations: OpenJev, a family of Jev-compatible decision servers on open-weight backends, and Decision 2.0. Then we’ll deploy both on Runpod Serverless, covering worker packaging, GPU selection, cold starts, and scale-to-zero. You’ll leave with a running decision endpoint and a clear sense of where these models fit in your stack.

​Bio

​Jessica Garson Beauchemin leads Developer Relations at Runpod. Previously a Senior Developer Advocate at Twitter and Elastic, she builds programs that turn developer feedback into product priorities and drive real adoption through documentation, advocacy, and community. Based in New York City, she is also a creative technologist and musician who performs live-coded noise music as Messica Arson, organizes RAID NYC, and speaks at conferences including DevRelCon, PyCon, and All Things Open.