Problem
A customer opens a ticket: they deleted an object by mistake and need it back, but it's just gone: there was never a prior version to recover. Around the same time, a different customer's backup job is hammering one narrow slice of the keyspace hard enough to bottleneck everyone else sharing that partition. Build an S3-like object storage service: PUT, GET, and DELETE for objects ranging from 1KB to 5GB, used as infrastructure by every other team's application. Design it from a blank canvas so object metadata (who owns what, permissions, size) is architecturally separate from the actual bytes, every byte is durable independent of which cost tier it's parked on, a delete is recoverable rather than instantly destructive, a large upload doesn't have to succeed or fail as one giant all-or-nothing request, a hot key range doesn't become a permanent bottleneck, and raw request throughput scales independently of how much data is actually stored.