ROUTEAI infra
JevPick
Abinashi singh@singh_abinashi
0:37JevPick is a tool that uses the Jev model to analyze a prompt's requirements and recommend the cheapest capable AI model for handling it, rather than defaulting to the most expensive option.
TakeClassification, scoring, routing.
Pattern↑ State to typed decisions for software
Demo, no signup: jevpick-production.up.railway.app
Most prompts don't need your most expensive model. But picking a cheaper one is a guess, so most of us just send everything to the top-tier model and pay for it on every call.
I came across TypeSafe AI's Jev model recently, and it's a different kind of thing from GPT, Claude or Gemini. It doesn't generate long responses. It makes structured decisions. Classification, scoring, routing.
That got me thinking: could Jev decide which model should handle a prompt, before that prompt ever reaches an expensive one?
So I built JevPick.
You give it a prompt. Jev reads what the task actually needs: difficulty, reasoning depth, production risk, how long the answer should be. JevPick checks those requirements against OpenAI, Anthropic and Google models, and recommends the cheapest one that should be capable of handling it. It recommends and prices a model; it doesn't answer your prompt.
Two examples from the demo:
A JSON extraction task → a cheap model, about 92% less than frontier pricing.
A race-condition debugging question → moves up to a top-tier model, and tells you the cheaper one was ruled out because its analysis capability was below what the task needed.
That second part is the bit I care about. A router that always picks the cheap model isn't routing, it's just a discount.
Still experimenting with the routing logic, but this was a fun way to actually build something with TypeSafe instead of only reading about it.
How are you handling model routing?
Related builds
