Why Benchmarking On-Device AI Turned Into Its Own Engineering Project

Benchmarking an LLM on a laptop is a solved problem. Doing the same on an Android phone turned into its own engineering project. Here is the tooling gap I ran into and the workarounds I had to build.