Sashidhar Guntury
About Me
I'm Sashidhar Guntury, an engineer who spends most of his time working on machine learning infrastructure and tooling.
A lot of my work sits at the intersection of model training, distributed systems, and observability. I like digging into problems at different levels of the stack—sometimes that's figuring out how to get training pipelines running smoothly on GPUs, other times it's making sure the right metrics and logs are in place to understand what's happening.
I also enjoy learning things from the ground up, whether that's exploring how attention works in plain Python before moving to optimized implementations, or tinkering with compiler internals through torch.compile and full graph compilations.
Areas I Work In
Infrastructure & Training
- • Training pipelines with PyTorch & PyTorch Lightning
- • GPU cluster workloads and monitoring
- • Grafana, Loki, dcgm-exporter
Model Internals
- • Tokenization and model internals
- • BPE, attention mechanisms
- • Plain Python to optimized implementations
Documentation & Tooling
- • Documentation workflows (MkDocs, MDX)
- • Python tooling and environment management
- • uv package management
Low-Level Exploration
- • Compiler-level experiments (pytorch compile)
- • Understanding systems from ground up
- • Performance optimization
Links
Personal Snapshot
Outside of the technical side, I like the idea that engineering often looks chaotic on the surface but has an underlying structure and I am interested in chasing that structure. The shirt in my photo is a good reminder of that. It may look trippy, but with the right perspective, patterns emerge.