Recruiter screen (30 mins) Standard background discussion. The recruiter verified alignment on compensation expectations, visa requirements, timeline, and briefly assessed my prior projects in computer vision and LLM fine-tuning.
Technical screening (60 mins — CoderPad + Video) Split evenly between coding and core machine learning fundamentals:
Coding: Implement custom Multi-Head Attention from scratch using PyTorch primitives (tensor operations, no torch.nn.MultiheadAttention). Focus was on batch matrix multiplications, correct shape tracking, and causal masking.
Theory: Walk through the vanishing gradient problem in deep networks, how layer normalization differs mathematically from batch normalization, and why AdamW is preferred over standard Adam for transformers with weight decay.
Dataset was noisy with missing values (~10% missing timestamps).
Required building a lightweight PyTorch Lightning / Hugging Face model, reporting precision-recall AUC curves, and providing a clean Dockerfile with clear inference benchmarks (latency under batch size 1 and 32).
The following metrics were computed from 1 interview experience for the AI21 Labs ML/CV/DL Engineering role in Israel.
AI21 Labs's interview process for their ML/CV/DL Engineering roles in Israel is incredibly easy as the vast majority of engineers get an offer after going through it.
Candidates reported having mixed feelings for AI21 Labs's ML/CV/DL Engineering interview process in Israel.