2026-09-19 · America/Los_Angeles · 社区动态 · #12

Tried using Jev for prompt observability and LLM routing

Been playing with Jev from TypeSafe AI and built two small experiments around it. One is a prompt oscilloscope that looks at a prompt while it’s being written and generates structured signals before the prompt reaches the main model. The other puts a small decision model in front of Claude as a routing layer. https://github.com/shubhangi013/prompt-oscilloscope https://github.com/shubhangi013/prune-review These are just personal experiments. Curious if others are using cheaper models for routing/gating before larger models, and what the economics look like in practice?

热度 44.6 / 100；排名与评分保留该期记录。