The presenter demonstrates Jev image analysisA vision-language model jointly processes images and language so it can describe, answer questions about or act on visual information. on support ticket screenshots, including classification and bounding boxes. The box answer always returns coordinates, so the presenter recommends checking whether the target is present separately.
A synthetic ticket experimentEvaluation is the systematic assessment of a system's behaviour or results against explicit criteria. illustrates reported accuracy, confidence thresholds and calibrationConfidence calibration measures whether an AI system's stated certainty matches how often its predictions or assumptions are actually correct.. The presenter describes human review for uncertain casesHuman-in-the-loop AI keeps a person involved in reviewing, correcting, approving, or guiding an AI system's work. and notes the limits of still-image inputs and request sizes.
Watch on YouTube
