Research Engineer
Atul Saxena
I build the environments models are trained in, and the tools that show what they learned.
- RL Systems
- Interpretability
- ML Infrastructure
Research Engineer
Atul Saxena
I build the environments models are trained in, and the tools that show what they learned.
Catching silent corruption in distributed GPU training
Real-time silent-error detection, weight-integrity monitoring and automatic recovery for PyTorch distributed training jobs.
A long training run that silently corrupts partway through costs the entire compute budget to discover — usually at evaluation, days later.
status: Active