Skip to main content
bash TV

Jev Explained: The Decision Model That Gives AI a Real Confidence Score

Agentic AI Institute

584 views22 Oct 2026

YouTube

Here’s how you can stay engaged and access more valuable content: 🔹 Sign up for the next session series: https://maven.com/lls/87e612 🔹 LinkedIn Community: https://www.linkedin.com/company/mahesh-ai-pm-community/ 🔹 Substack : https://myaicommunity.substack.com/ 🔹 Sign up for Free Learning Resources: https://forms.gle/sRNjXnsurNxNAUQW7 🔹 AI PM Certification: https://maven.com/mahesh-yadav/genaipm 🔹 AI PM Interview Bootcamp: https://maven.com/mahesh-yadav/ai-pm-interview-prep-bootcamp 🔹 Speak to Our Mentors: Book a call Here: https://calendly.com/d/ct2s-3s2-n89/1-1-with-team-mahesh?month=2026-10&utm_medium=youtube&utm_campaign=description_campaign LLMs are trained to generate. Automation needs decisions. And when an LLM gets a decision wrong, its confidence is still 99%. In this Lightning Session, Mahesh Yadav decodes Jev, a decision model, from a product perspective: what it is, what it changes, and where to apply it in your own product. What we cover: → How we got here: small language models, large language models, RLHF and RLVR → Why generation models cannot give you a reliable confidence score → RLCD: reinforcement learning with calibrated decisions → The three decision types: choice, score and yes/no → A contract workflow with 8 model calls, 6 of which are decisions → Using confidence thresholds to route uncertain cases to a human → Hands-on labs: the Jev playground, and Jev as a judge for evals → Eval results: Jev vs a frontier LLM vs small LLM judges on speed and cost Key takeaways for AI PMs: → Map your user flow and mark every decision, separate from generation → Set a confidence threshold for each decision. Too high and users get nagged. Too low and errors flow downstream → Break your evals into yes/no questions so a decision model can run them Jev cannot explain, reason or summarize. It is built to make decisions with calibrated confidence, and that is the point. Session page: https://maven.com/p/b5dd86/evaluate-ai-with-jev-when-classification-beats-an-llm Join the Lightning Session series: https://maven.com/lls/87e612 Free Lightning Sessions are Step 0 for anyone moving into AI product management. #AIPM #AgenticAI #AIEvals #DecisionModels #Jev

Join the discussion

Sign in to join the discussion

Sign in