Warner Bros. Discovery · Primly Community

Warner Bros. Discovery senior / L5 system design interview: what to expect and what actually mattered

quietquit_quincy · 4 replies

interviewed for a senior / L5-equivalent role on the platform engineering side at WBD (the Max streaming backend) a few months back. the system design round was the most interesting part of the loop, so writing this up specifically.

the prompt they gave me was something like: design a video content metadata service that supports global search and recommendations at scale. not verbatim but close enough. classic media-company problem. you're designing for a catalog of millions of titles, multi-region delivery, real-time personalization signals, and editorial override capability (which is a very WBD thing to care about).

what they were actually evaluating:

I asked the interviewer after and they confirmed the rubric is roughly: can you drive the conversation and ask the right clarifying questions upfront? do you understand the tradeoffs between consistency, availability, latency at scale? can you talk sensibly about caching layers (CDN, in-memory, read-through) without over-engineering? do you know when to use SQL vs. NoSQL and WHY for THIS problem, not in the abstract?

I leaned on a Cassandra approach for the metadata store with Elasticsearch for search indexing, Redis for hot path caching. they pushed back on my elasticsearch choice and asked how I'd handle index lag for newly ingested content. that was the key moment. they wanted to see how I handled the pushback and whether I could reason through the tradeoff in real time.

senior vs. staff signal: I was told later that the differentiator between a strong senior and a staff hire at WBD is whether you proactively mention operational concerns. latency SLOs, runbook gaps, observability. I mentioned adding structured tracing and a degraded-mode fallback and the interviewer visibly lit up.

format: 60 minutes. first 5 on setup and clarifications, ~40 on designing and whiteboarding (they used a shared miro-like tool), last 15 on their questions and drill-downs.

if you're preparing: do one practice run on a media-domain problem specifically. video metadata, content recommendation, search indexing. it's different from the payments or ride-sharing prompts you usually see, and WBD interviewers clearly care that you understand their domain a little.

4 replies

ae_andre

the elasticsearch lag question is a classic and interviewers love it when candidates have actually hit it in prod. i've used a dual-write pattern with a pending index queue that gets reconciled async. did you go that direction or something else?

sdr_sky

i went with a change-data-capture approach from the primary store into a Kafka topic consumed by the indexer. mentioned that for newly ingested content you'd flag it with a "pending" state visible in search rather than hiding it entirely. interviewer seemed to find that reasonable.

ml_mike

did the recommendations piece come up at all, or did you keep it out of scope? curious whether WBD wants SWEs to have opinions on the rec system architecture or if they treat that as a separate ML infra concern.

director_dee

the operational concerns signal is spot-on for senior vs staff leveling. I use the same bar when I'm in the room. if someone designs a beautiful system but never mentions "how will we debug this at 2am" they're a strong senior, not staff. the instinct to think about oncall burden is earned, not taught.