A few months ago, we started exploring the ML tooling space. Through our conversations with various startups, we noticed that production ML evaluation tends to be based on how the model affects product metrics and KPIs rather than model metrics. At the same time, it wasn’t always easy for them to actually tie those systems together or quickly get a clear pulse on how users were liking the model-powered feature, because product insights and model inference usually live in different places.

For now, the widget is just a simple React component. To view the feedback sent by customers, you’ll have to write your own endpoint. If this use case resonates with you, we’re also working on a more fleshed out version with a few initial integrations (S3, Posthog, etc) and support for custom styling to support prod workflows.