@sakshityagi
About
I build AI systems for places where being wrong has consequences and I've worked the full stack to get there, from GPU interconnect up to the production application.
As a Solutions Architect at NVIDIA, I designed and shipped end-to-end AI systems: distributed training and inference on multi-GPU clusters, model and pipeline optimization with TensorRT and Triton, and the architecture work of taking systems from prototype to production. The deployments spanned healthcare and life sciences through robotics and enterprise but the through-line was always the same: making AI actually run, reliably and affordably, at scale.
Today I'm a Senior Applied Scientist working on LLMs in a domain where a confident wrong answer isn't a cute demo bug , it's the whole problem.
I write about production generative AI on hard mode: distributed training, evaluation and trust, inference economics, and the engineering that separates a shipped feature from a magic trick. Generalized lessons from real work. No hype.
The model was never the hard part. Everything around it is.
Available for
Nothing here yet.
