Heavy emphasis on practical coding under time pressure and on reasoning out loud about safety trade-offs.

Applied ML systems, inference-time performance work, and a debugging round on unfamiliar code.

Integration-style problems on a real API surface, plus correctness questions around money and idempotency.

Distributed data processing, query planning intuition, and one deep round on a system you have actually operated.

Storage and concurrency internals, with a strong preference for candidates who can reason about cost.

Geospatial and real-time matching problems, then a design round on something with a live dispatch loop.

Fewer puzzles, more judgement: architecture reviews, failure post-mortems, and blunt culture conversations.

Rendering performance, CRDT-flavoured collaboration questions, and a hands-on front-end round.

Retrieval pipelines, ranking evaluation, and shipping speed — expect to write working code quickly.

Product-minded backend work: schema design, ledger correctness, and a pragmatic take-home style round.

Game-loop performance, memory budgets, and network topology for many concurrent clients.

Training-infrastructure questions, GPU utilisation reasoning, and low-level optimisation.