Overview & result
Notra is a GEO tool that asks ChatGPT, Claude and Gemini the questions buyers ask and shows whether a brand appears. Jev is not mentioned in the main README; the repository's code and agent docs use the typesafe-ai/jev evaluation model for the chat router that picks a model per turn, for typed sentiment and position evaluation next to the GEO judge LLM, and for a feedback classifier.
From the original post
Notra is a modern GEO tool that asks ChatGPT, Claude and Gemini the questions your buyers ask. See if you show up, who shows up instead and how to fix it.
How Jev fits in the loop
- Notra runs buyer questions across AI engines and stores the answers.
- For chat, the root agent picks its model per turn by asking the typesafe-ai/jev evaluation model.
- In GEO checks, a typed Jev evaluation of sentiment and position runs beside the judge LLM.
- An environment flag switches these classifiers back to LLMs.
How to reproduce
- Inspect the original showcase and source material at https://github.com/usenotra/notra
- Verify the bounded prompt and input candidate schema configured for Jev
- Benchmark decision latency and classification accuracy against baseline models
Why this build matters
It shows Jev used inside a shipping product as a router and evaluator, not as the product's headline feature.
Limitations
- Jev is not mentioned in the main README; these details come from the repository's code and agent docs.
- The repository gives no measurements of the Jev classifiers.
- Whether a mention exists is decided deterministically; Jev covers sentiment and position only.