Decision Models and autoresearch: getting value in your product, using Jev to analyze reviews

Learn Decision Models by building a service that labels reviews by topics and sentiments

You’ll learn what Decision Models are: a new class of AI models that don’t write text but choose an answer from a set of predefined options, dozens of times faster and cheaper than LLMs.

We’ll look at which tasks they handle better than LLMs, how to have a coding agent tune your prompt using the autoresearch approach, and how to build a cascade in which the Decision Model makes the confident decisions and hands the uncertain ones to an LLM.

You’ll put all of this into practice by analyzing 7,000 reviews with the Jev model. By the end, you’ll have a service that labels every review across 15 topics and their sentiment, an admin panel for experimenting with prompts and models, and analytics showing what users praise, what they complain about, and which topics are on the rise.

We’ll also show you how to find tasks for Decision Models in your own product and how to audit the LLM pipelines you already run.

By the end of this case you'll have a working product and a set of reusable skills:

working with Decision Models: Jev and OpenAI’s Decisions API
autoresearch: a coding agent runs experiments and tunes the prompt against a metric
labeling with LLM-as-a-judge and evaluating quality on training and test datasets
cascading a Decision Model with an LLM and choosing a pipeline by quality and cost
auditing the LLM pipelines in your product and finding tasks for Decision Models

How do I get access to the program?

This case is part of the GoPractice “Build with AI” subscription. You can start right now.