AT&T · Primly Community

AT&T senior / L5 system design interview, what to expect: my notes from the loop

sre_sol · 5 replies

Just wrapped the full loop for a senior SWE / tech lead adjacent role at AT&T's technology and operations division. The system design round deserves its own write-up because it's different from what most people prep for.

First: AT&T doesn't use L5 as a label. Their bands are something like SWE I / II / III and then Principal. The role I interviewed for was their SWE III level, which from comp and scope is roughly equivalent to senior / L5 at Big Tech.

The system design format: One hour, one interviewer (a principal engineer). He opened by saying he wanted to discuss a problem relevant to what the team actually does. He gave me a design problem involving a large-scale billing event pipeline. Think: millions of events per minute, subscribers generating usage data across 5G and fiber, needing real-time aggregation for billing and anomaly detection.

This wasn't a generic "design Twitter" type question. It was domain-specific and they expected me to at least ask smart questions about the telecom context, not pretend I'm designing for a web startup.

What he cared about most: Fault tolerance and exactly-once delivery (billing errors are really bad for a carrier) How I'd handle backpressure when consumers fall behind Data partitioning strategy and why Trade-offs on consistency vs. availability, and I had to be specific about which parts of the pipeline could tolerate eventual consistency

I used Kafka for the event backbone, Flink for stream processing, and Postgres + a columnar store for the aggregation layer. He challenged every major choice. Not in an adversarial way, more like an architect reviewing an actual design.

What I wish I'd prepared more: Telecom-specific data volumes and SLA expectations. I knew distributed systems but I kept having to say "I'm not sure what typical AT&T-scale numbers look like" which was fine, but knowing rough orders of magnitude (like: they have 200+ million subscribers, events per session, etc.) would have let me anchor the design better.

I got through to an offer. System design felt like the heaviest weight in the decision.

5 replies

infra_ines

Did he go deep on Kafka internals or was it more high-level? I always get nervous about how far to go on infra choices.

sre_sol

He asked about consumer group offsets and what happens when a consumer restarts mid-batch. So yeah, one level below surface. I'd say know your tech choices well enough to defend the failure modes.

backend_bekah

Was this fully virtual or did you go to a Dallas campus for it?

sre_sol

Fully virtual. They said onsite is an option but they've mostly standardized on video for the interview rounds.

hardware_hugo

Good write-up. The billing domain angle makes sense for a carrier. Most candidates prep for e-commerce or social scale problems and then get caught flat-footed when the domain is telecom infra. AT&T is not Netflix.