Problem
A team ships a 2-point accuracy win and gets paged at 2am: the new model is roughly 3x too large for the serving fleet, and p99 latency tripled the moment it went live. The release process had checked exactly one thing; accuracy; and nothing else. Nobody's buying new hardware to fit this model; the serving fleet is what it is, handling roughly 500 requests/second at a p99 budget of 150ms, same as before the release.
Required components
Click to add a component to the canvas.
Additional components
Browse by category. Click or drag a component onto the canvas.