{"type":"video","version":"1.0","html":"<iframe src=\"https://www.loom.com/embed/91b07743e50a4179b87487c6ab6ef5d7\" frameborder=\"0\" width=\"1724\" height=\"1293\" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>","height":1293,"width":1724,"provider_name":"Loom","provider_url":"https://www.loom.com","thumbnail_height":1293,"thumbnail_width":1724,"thumbnail_url":"https://cdn.loom.com/sessions/thumbnails/91b07743e50a4179b87487c6ab6ef5d7-06cf37806fa5b61d.gif","duration":134.85,"title":"Typesafe app &amp; Jev AI model in Make [DEMO]","description":"This Loom introduces Jev by Typesafe, a decision-making AI model designed for fast, cheap, and low-error judgments, particularly for evaluating email threats before other LLMs respond. The speaker routes received emails into Jeff, maps email content into a threat classification question, and labels outputs such as pass, flag, or threat. In a test, Jeff scored data exfiltration at 0.75 and credential requests at 0.99 yes, resulting in a flag and a Slack message for screening. The sample email subject was \"need a production webhook key before end of day,\" and the speaker concludes the decision was likely correct."}