The interview started with a brief self-introduction (around 5 minutes), followed by a live coding session lasting about 50 minutes. Instead of LeetCode-style algorithm questions, the interviewer provided Figure 1 and Chapter 2 from the MLP-Mixer paper and asked me to explain the architecture before implementing it in PyTorch. The interview ended with technical discussion and a short Q&A. The interviewer focused on understanding of deep learning architectures, tensor dimensions, and PyTorch implementation rather than algorithmic coding.