Live demo
IntermediateTry it: deterministic triage for member inquiries
This is the working version of our Laya triage tutorial, running right here in your browser. Paste an inquiry, watch it get scored against four departments, and see the confidence gate catch the ambiguous ones. Same input, same output — every time.
What you just ran
A fixed set of four categories, each with a keyword list. Every keyword hit scores one point; the top scorer wins — but only if it clears the confidence gate. Below the gate, the inquiry goes to a human instead of being guessed at. There is no sampling, no temperature, no model: the same text always produces the same scores. That is the whole point of the tutorial this demo comes from.
- Fixed categories — the departments you define, not a model improvising labels.
- Confidence gate — ambiguous inquiries route to a person. The gate refusing to guess is the feature.
- Deterministic — hit "run it twice" and compare. Reruns agree, which is exactly what LLM sampling can't promise.
The full tutorial replaces keyword scoring with Laya, a local decision model, for the harder cases — same gate, same idea, still on your own machine. Read the tutorial.
Request the starter kit
The demo above runs a simplified version in your browser. The starter kit is the real thing: a single Python file implementing this exact logic for your own machine, plus the human-review worksheet.
Want the files behind this lab, sample dataset included? Email ai@trihelix.co and we will send them your way.