First I talked with the recruiter, then they gave me a take-home with one week to finish. The take-home was optimizing a benchmark testing setup — there's a POST endpoint and a GET endpoint, using S3 as the database. The version they hand you originally reads the whole JSON out every time and uses string find to look for the offset. The overall speed of the POST was about 20+ seconds, and the goal was to optimize it down to around 5-7 seconds. They require you to write a report with data analysis, and to write up the differences and reasoning behind each optimization choice you made.
Then straight into onsite — 3 rounds, all on the same day.
30 minutes, Hiring Manager — mainly talked about why I wanted to work at LangChain, my biggest challenge, and my most complex project.
30 minutes, code review — went back over the take-home. The interviewer had me explain the decision I changed the most, how I judged that it was the right decision, and how I'd redesign it if I did it over.
60 minutes, system design — design LangSmith. How would you help customers track tokens and metrics, and speed, and how would you handle and design an alerting layer on top of that. Alerts need to cover things like a condition occurring N times within M minutes, and once the threshold is passed, making sure the customer has actually been notified — basically design a product like Datadog. The frontend needs to support a dashboard, monthly billing, and alert controls. At the end they also asked about anomaly detection — how you'd do it — and how to handle late-arriving events and duplicate events.
Discussion
Loading comments…