Jev as a Decision Layer: 8 Agent Use Cases + Live Demos | Youssef Hosni

Most of your agent's LLM calls don't need an LLM. They need one fast, reliable decision. In this full workshop, I show how to use Jev as a decision layer for AI applications and agent workflows: what it is, how it works, 8 practical use cases, and live demos comparing it to an LLM. In the model-routing benchmark, Jev picked the right route in 36/36 cases (GPT-4.1 mini got 35/36), ran about 2.5–3× faster, and cost a fraction as much per decision. What you'll learn: • What Jev is and why it fills a real gap in agentic workflows • How a Jev request works: state, questions, criteria, and outputs • The 3 decision primitives: choice, score, and null (yes/no) • How confidence is computed and when to escalate to an LLM or a human • 8 use cases: model routing, guardrails, tool-call gating, inbox triage, RAG re-ranking, LLM evaluation, bulk labeling and real-time control • Live code demos: Jev vs GPT-4.1 mini on latency, accuracy, and cost • How to evaluate a decision layer inside your own agent Chapters 0:00 Welcome & workshop agenda 1:38 What is Jev? Background and why it matters 4:07 Jev as a multi-class classifier 5:44 How Jev works: state, questions & the request format 12:54 Decision primitives: choice, score and null 18:05 Jev vs an LLM: speed & structured outputs 21:57 Confidence: when to trust a decision 24:26 Overview of the 8 use cases 24:53 Q&A: Jev vs LLMs, cost, latency & accuracy 31:26 Use case 1: Model routing 35:57 Use case 2: Guardrails & prompt-injection detection 39:25 Use case 3: Tool-call gating 41:39 Use case 4: Inbox triage 43:22 Use case 5: RAG re-ranking 46:20 Use case 6: LLM evaluation (+ Q&A) 53:03 Use case 7: Bulk labeling 54:02 Use case 8: Real-time control (games & control loops) 56:00 Demo 1: Model routing, Jev vs GPT-4.1 mini 59:43 Benchmark results: latency, accuracy & cost 1:01:35 Demo 2: Inbox triage 1:02:54 How to evaluate Jev inside your agent 1:04:19 Q&A: many labels, human-in-the-loop, how Jev is trained 1:12:23 Wrap-up Codes: https://github.com/To-Data-Beyond/Jev... Resources mentioned: • Jev docs (TypeSafe): https://docs.typesafe.ai/models#custo... • Kev, an open decision model: https://huggingface.co/jaredpalmer/ke... • Laya, an open decision model: https://github.com/NandhaKishorM/laya 📬 Want deeper dives like this? Subscribe to To Data & Beyond, my newsletter on AI, LLMs, and agents: https://todatabeyond.substack.com/ Instructor Youssef Hosni is an AI & Machine Learning Engineer at Solita and the founder of To Data & Beyond, where he teaches practical AI and Generative AI through articles, a newsletter, books, and live workshops. #AIAgents #LLM #Jev #AgenticAI #GenerativeAI #MachineLearning #RAG #AIEngineering