SRE / Platform Engineer. I make infrastructure reliable, and lately I'm figuring out how to do the same for GPUs and inference -- in the open.
- frit -- how reliably can a single GPU run the full inference stack? vLLM, DCGM, k3s, SLOs, chaos -- finding out on real hardware.
- platform-zero -- a whole SRE platform from a blank Linux box: GitOps, secrets, observability, IaC.
- ducat -- one cost view across cloud, SaaS, and AI spend, normalized to the FOCUS spec and exported to Prometheus/Grafana.
- florin -- a portable database workbench for VS Code; your connections and saved queries sync to a Git repo you own.
- Chipping in on NVIDIA's GPU stack -- dcgm-exporter, gpu-operator.
Resume · 75asu.pages.dev · linkedin.com/in/75asu · asutosh.pda@gmail.com


