Speed up object-backed APIs, then design LLM observability

Two parts. First, make slow API endpoints whose data lives in stored objects faster. Then design an observability platform for LLM applications: capturing traces of model calls and agent steps, storing and querying them at scale, and helping developers debug and evaluate their apps.