How are you vetting ML engineers lately

Anyone else seeing take-home fatigue with senior candidates? I’m sourcing on GitHub and Kaggle with about a 28% reply rate, and I swapped HackerRank for a 45-minute repo walkthrough + live pair to probe data leakage and GPU memory profiling — curious what’s giving you signal right now.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌⁠​‍‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠‌‌⁠⁠‌⁠‌​‌‍⁠⁠‌⁠​​‌‍‍‌‌‍​⁠​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​‍​‍‌‍⁠‍‌‍‌‌‌⁠‌⁠​‍​‍​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‌​⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍‌‍‍‌‌⁠‌‌‌‍‍‌‌‍​‍‌​‍⁠‌‍‍​‌​⁠‍‌⁠‍‌‌‍‍‍‌‍⁠‌‌‍⁠​‌⁠​⁠‌​⁠⁠‌‍‌​​⁠‍‌​⁠​​​‍​‍‌⁠⁠‌​​

But your ‘45-minute repo walkthrough’ works; add 5-minute PyTorch profiler trace read for OOMs.

‌⁠‍⁠​‍​‍‌⁠‌​​‍​‍​⁠‍‍​‍​‍‌‍‌⁠‌⁠​‍‌‍‍‍​‍​‍​‍⁠​​‍​‍‌‍‍⁠​‍​‍​⁠‍‍​‍​‍‌⁠​‍‌‍‌‌‌⁠​​‌‍⁠​‌⁠‍‌​‍​‍​‍⁠​​‍​‍‌‍‍‌‌‍‌​​‍​‍​⁠‍‍​⁠​‍​⁠​​​⁠‌⁠​⁠​‍​⁠‍‌​‍⁠​​‍​‍‌‍‌​​‍​‍​⁠‍‍​‍​‍​⁠​‍​⁠​​​⁠​‍​⁠‌‌​⁠​‌​⁠​‍​⁠​‍​⁠‌​​‍​‍​‍⁠​​‍​‍‌‍‍​​‍​‍​⁠‍‍​‍​‍​⁠‌‌‌‍‌‍​⁠​​‌​‌‍‌‍‌‍‌⁠‍​‌​​‌‌‍‍⁠​⁠‌‍‌⁠‌‍‌‌‍‌‌‍‍⁠‌​⁠​‌‍⁠‍‌‌‌‌‌​‌‌​‍​‍‌⁠⁠‌