Hi HN - I built this agent to make it easier to get feature distribution metrics into a Prometheus-Grafana observability stack when deploying with Tensorflow Serving.
The pain point this addresses is that TF Serving's default /metrics endpoint only exposes performance metrics (request size, request counts) but not metrics on the features or predictions, which is often interesting and valuable to monitor.
The pain point this addresses is that TF Serving's default /metrics endpoint only exposes performance metrics (request size, request counts) but not metrics on the features or predictions, which is often interesting and valuable to monitor.