Machine Learning Engineer · Onsite · Multiple Types

Amazon — Machine Learning Engineer ✅ Passed

Level: Senior-Level

Round: Onsite · Type: Multiple Types · Difficulty: 7/10 · Duration: 180 min · Interviewer: Friendly

Topics: Large Language Models, System Design, Behavioral Questions, Algorithms, Transformers, Model of Experts, Parallelism

Location: Seattle, WA, US

Interview date: 2025-12-01

Summary

Round 1: Coding

Question: I was asked about the details of my previous LLM projects and fundamental LLM knowledge, such as Transformer and MoE structures, collective communication, and parallelism. Then I was asked to solve a coding problem.

Round 2: Leadership Principles (LP)

Question: I was asked about project details and leadership principles.

Round 3: System Design

Question: I was asked to design an LLM quality validation system.

Details

The onsite interview consisted of three rounds.

In the system design round, I addressed the LLM quality validation system design from three perspectives:

1. Validation at different levels: layer-by-layer, tensor, logits, tokens.
2. Using different methods: llmaaj, loss function, human annotation.
3. End-to-end system: Users issue asynchronous commands to launch EC2, start Triton server and VLLM, run inference, collect LLM responses, upload to S3, and validate using the above methods.

The coding question I received was:

• LeetCode 1642, meeting rooms problems, solved using Max-heap

LeetCode similar: LeetCode 1642