A recruiter call, a 60-minute coding screen, sometimes a take-home for senior or infrastructure roles, then coding on GPU-cluster style problems, an applied ML systems round (attention, streaming, batching, KV cache), system design for senior roles, and a final talk with a leader.
Diagnose a slow Linux server: CPU, disk, network and logsCodingA troubleshooting round in several scenarios. For each, say which tools you reach for, what the numbers tell you and how you get from the host down to the cause. 1. CPU looks… Split a file iterator into balanced line-aligned partsCodingSplit a large file into balanced parts for parallel workers without cutting lines in half: each part holds whole lines and the parts stay close in size. Reported on 1point3acres'… Remove a GPU node and reschedule its podsCodingGiven a cluster of GPU nodes running pods, drain one node and reschedule its pods onto the remaining nodes respecting their resource requirements. Reported on 1point3acres'… Design a GPU scheduling platformAI systemShare a large GPU cluster among research and production teams: queue training and inference jobs, enforce quotas and priorities, place jobs on topology-aware hardware, and preempt… Temporal key-value store with point-in-time reads and restoreCodingBuild an in-memory key-value store that keeps every version of every key, so you can ask what a key held at any point in time. Implement the TemporalKVStore class: -… Detect and break cycles in pod dependenciesCodingYou have many pods, and a pod may depend on other pods. 1. Dependencies are directed. Report whether they contain a cycle, much like checking whether a course schedule can be… Find eligible GPU nodes for a podCodingFilter a fleet of GPU nodes down to the ones that can host a pod with given CPU, memory and GPU requirements. Reported on 1point3acres' Together AI bank; reconstructed in our own…