Low-end Android devices expose performance problems that flagship phones often hide. Small inefficiencies in startup, memory usage, caching, or background work can quickly turn into ANRs and a frustrating user experience at scale. In this talk, I’ll walk through how we approached a performance overhaul at Super, from local profiling to production telemetry. We’ll use memory dumps, Perfetto traces, flamegraphs, Macrobenchmark, and microbenchmarks to identify CPU and memory bottlenecks, then look at the trade-offs behind architectural choices such as on-disk caching, in-memory caching, and memory-efficient data structures. We’ll also cover how to ship risky performance changes safely using feature flags, and how to validate their real-world impact with production observability. By the end, you’ll have a practical blueprint for finding Android performance bloat, reducing ANRs on constrained devices, and proving that your optimizations actually work in production.