~/ shubham tatvamasi
GPU sharing, LLM serving and Kubernetes platforms, from a single node to fleets of clusters, engineered so models run fast, GPUs never sit idle and nothing goes dark.
Most people meet AI through a chat box. I live one layer down, where tokens become GPU cycles, GPU cycles become heat, and a single misrouted request can idle a rack. My work is making that layer boring in the best way: declarative, observable, self-healing, and secure enough that nobody has to think about it.
Making the whole stack, from the GPU die to the API gateway, behave like one calm, observable, self-healing system.
One control plane for models served through KServe, NVIDIA Dynamo, NIM, vLLM and SGLang. Model-aware routing through Envoy AI Gateway and llm-d, budgets and keys via LiteLLM, GuideLLM benchmarks, and autoscaling that adds replicas before it adds GPUs.
Telling allocated apart from actually computing. DRA with MIG, MPS and time-slicing, Kueue and KAI queues, and NVIDIA and AMD telemetry that catches idle, faulty and oversubscribed GPUs.
Controllers, CRDs and operators in Go. Multi-cluster networking, tenant isolation and HA storage that heals itself.
Flux CD reconciling 80+ clusters. If it isn't in Git, it doesn't exist.
Post-quantum key exchange, signatures and certificate lifecycles, automated as Kubernetes-native resources.
Encrypted tunnels from 900+ devices behind carrier-grade NAT to private and public clouds. Fleets that stay reachable, observable and patched no matter where they live.
From web apps to edge fleets to GPU fleets. The workloads changed; the job of making them run calmly didn't.
Architecting the GPU & AI module for Portainer Enterprise: cross-vendor GPU observability, Kubernetes DRA sharing with MIG, MPS and time-slicing, Kueue and KAI Scheduler queues, and one control plane for models served through KServe, NVIDIA Dynamo, NIM, vLLM and SGLang.
End-to-end AWS architecture for an AI startup’s SageMaker platform, Spot migration that cut compute cost by ~30%, and a multi-cluster EKS platform on Gateway API and VPC Lattice.
Post-quantum encrypted connectivity for 900+ field-deployed edge devices in a government OpenStack datacenter, with WireGuard and FRP tunnels through carrier-grade NAT. Created QubeSec along the way.
Earned a TSC seat through contributions alone, and designed Magma’s first private-cloud deployment path, now part of the official install.
Founded the Cloud Native Security India community under CNCF and tripled contributor participation.
Led technical direction and release automation for Kubernetes-based telecom platforms in production.
Go controllers for on-demand academic microsites, a hybrid OpenStack, AWS, Azure and Proxmox migration, and 200+ websites behind a single public IP.
A multi-tenant ML competition platform on Kubernetes, Istio and Kubeflow, built to run untrusted customer workloads in isolation.
Where it started: React and Node on AWS, then smart contracts, Hyperledger and Raspberry Pi mining clusters.
A short operating manual for the systems I build, and the way I build them.
Every cluster, model and secret should be reproducible from a commit. If a human has to remember it, it will be forgotten.
Allocated isn’t utilised. Ready isn’t responsive. I chase the metric that tells the truth, not the one that looks good on a dashboard.
Replicas before nodes, caches before GPUs, routing before hardware. Efficiency is a design decision, not a cost report.
Harvest-now, decrypt-later is real. Post-quantum crypto belongs in today’s pipelines, not tomorrow’s roadmap.
The best infrastructure ideas get sharper in public. Ship it, document it, let others break it.
A tiny, very real-feeling terminal. Type help to look around, or try nvidia-smi.