Context
A high-growth SaaS platform was missing p95 targets during peak hours. Feature work couldn't stop, and security debt was accumulating.
Approach
- •Modular-monolith plan; dual-run/blue-green cutovers
- •Distributed tracing on hot endpoints; DB index & cache tuning
- •Error budgets + observability with SLOs
- •No feature freeze; release guardrails