Went through the KPMG system design round last month for a senior software engineer role (they don't use L-numbers publicly but internally it maps to something like a senior IC, call it L5-adjacent). Sharing because I couldn't find anything specific before my interview.
The round is 50 minutes. You get one problem. Mine was: design a real-time audit logging system for financial transactions across multiple client tenants. Classic Big 4 flavor, the use case is always enterprise / compliance / financial services adjacent.
What they actually cared about: Data ingestion at scale: how do you handle thousands of events per second from multiple sources. I talked through Kafka, partitioning by tenant ID, consumer groups. Storage tier: they pushed me on why a time-series DB vs. relational. I talked through retention policies, hot vs. cold data, query patterns. Multi-tenancy isolation: this came up hard. How do you make sure Tenant A's audit logs can't be queried by Tenant B. Schema-per-tenant vs. row-level security, tradeoffs. Reliability and durability: what happens if a node goes down mid-write. They were looking for at-least-once vs. exactly-once delivery awareness.
They did NOT ask me to design a URL shortener or a Twitter feed. It was very focused on the consulting domain, which makes sense.
Your interviewer was a senior engineer who kept asking follow-up questions. It felt collaborative, not gotcha-style. When I didn't know something about a specific AWS service behavior, I just said 'I'd have to verify the exact SLA guarantees' and they moved on without penalizing it.
Time management: I spent too long on ingestion and had to speed through monitoring. Leave at least 10 minutes for observability, it came up at the end and I was rushed.
Prep that actually helped: designing data-intensive systems with compliance constraints in mind, not generic FAANG system design templates.