Machine Learning Engineer · Onsite · Multiple Types

Amazon — Machine Learning Engineer ❌ Failed

Level: Intern

Round: Onsite · Type: Multiple Types · Difficulty: 6/10 · Duration: 120 min · Interviewer: Neutral

Topics: Machine Learning, Transformers, System Design, Behavioral Questions, Adam, PPO, VQVQE, RoPE, CLIP

Location: Seattle, WA, US

Interview date: 2026-02-10

Got offer: False

Summary

Round 1: Technical and Behavioral

Question: Questions about Transformer architecture, advantages of Transformers, differences between Adam and AdamW, PPO concepts, VQVQE training challenges, RoPE formula, and differences between CLIP and Qwen-VL architectures. Also included behavioral questions.

Round 2: Technical, System Design, and Behavioral

Question: Questions about machine learning concepts, system design problems, and behavioral questions.

Details

In the first round, I was asked about the following:

• Transformer structure
• Advantages of Transformers compared to other models
• Differences between Adam and AdamW
• PPO concepts
• VQVQE training challenges
• RoPE formula
• Differences between CLIP and Qwen-VL architectures
• Handwritten implementation of InfoNCE in PyTorch

I was also asked behavioral questions about incomplete information and delivering before a deadline.