Show HN: I used Jev to build an off-topic detector for meetings (maxharrison.me)

A developer has created an "off-topic detector" for meetings using Jev, a recently released decision model by TypeSafe AI. This tool, called the Offtrackometer, aims to signal when discussions stray from meeting objectives, addressing a common issue many professionals face. With an initial demo available for free, the project evolved from previous attempts with other models that lacked reliability, showcasing Jev’s ability to provide quick and accurate classifications within defined contexts. The Offtrackometer utilizes local speech recognition from Chrome and the Clef-flash decision model on Cloudflare to analyze meeting discussions. It assesses whether conversations align with preset goals, returning results in real-time as a percentage likelihood of being off-topic. This practical application highlights Jev's efficacy over autoregressive LLMs, offering faster response times and cost-efficient classification. Future plans include gathering evaluations from real meeting data to improve the tool's accuracy and exploring the feasibility of running decision models locally in a browser. The developer welcomes feedback and potential use cases, marking this project as a stepping stone in integrating decision models into everyday workflows.