Free Lesson
Can a Decision Model Think Like a Search Engineer?
60 min
Oct 28, 2026 11:30 AM
By continuing, you agree to Maven's Terms and Privacy Policy.
What you'll learn
How tuning search is full of somewhat automatable judgments
How Relevan picks relevance features per customer index, and when heuristics & LLM calls with parsing issues fall short.
SystemOne Models: Frontier-LLM accuracy at ~14x the speed
See Jev, Laya, and Clef benchmarks. One matched Claude Opus 5 on all config decisions at 0.16s vs. 2.27s per answer.
Let the model’s confidence tell humans where to look
How calibrated probabilities let reviewers skip the easy fields, and how to run the open eval against your own model.
Why decision models should now inform how we build retrieval
Agentic search, automated relevance tuning, relevance judgments. What new opportunities do these new models open up.
Why this topic matters
Every search index hides judgment calls. For example, should a name field match misspellings by similar sound? Enable this on the wrong field and you add bad matches. But new decision models return typed answers with calibrated confidence, so can they make those calls? David Tippett, founder at Relevan, tested Jev, Laya, & Clef. It turns out the answer is yes!... in many cases. Hear experiments and benchmarks firsthand, and bring your questions.
