Before you press play
What you’ll find in the video
- Jev evaluates state text and returns structured probabilistic scores rather than generative free-form conversational text.
- Requests define questions using schema types such as Noul for yes/no probability scoring and choice for multiclass criteria evaluation.
- A single HTTP request can evaluate multiple questions simultaneously, returning normalized decision probabilities and token usage metadata.
Worth knowing
Gemini-assisted video/transcript review. The author tests an informal sample prompt via HTTP in C#, not a controlled benchmark, and does not evaluate production error handling or calibration.