GPU-Aware Kubernetes: A Beginner's Guide to DRA, Kueue and NVIDIA's Tools
A beginner's guide to GPU-aware Kubernetes: how DRA picks the right GPU, how Kueue starts a job only when all of it fits, and where the NVIDIA GPU Operator and Run:ai fit in.
Deep dives into systems engineering, custom operators, and private LLM platforms.
Writing about custom Kubernetes operators, low level interfaces, and private LLM infrastructure setups.
Beyond One GPU: How LLM Inference Scales Across Clusters
How LLM inference scales past one GPU: data, pipeline, tensor and expert parallelism, prefill/decode disaggregation, and running it all on Kubernetes.
After Ingress-NGINX: What Actually Replaces the Retired Controller
Field notes on Kubernetes networking: what really happened to ingress-nginx, what Gateway API and ingress2gateway do and don’t solve, and what’s genuinely still missing before you cut over.
Serverless-Bend: Massively Parallel APIs in a Single Command
Exposing the parallel-by-default execution of Bend as highly scalable, multi-cloud serverless endpoints.
An Introduction to Bend: High-Level Parallelism for All
A new programming language that makes parallel computing on CPUs and GPUs feel like writing Python, without the complexity of C++ or CUDA.
Building CloudDB: A Self Hosted Database Platform for Kubernetes
Running databases on Kubernetes has a reputation for being painful. I built CloudDB to change that: a Kubernetes native, self hosted DBaaS control plane with a React dashboard and a Go API that wraps the Zalando Postgres Operator. Here is what I built, why, and where it is heading.
MLOps: Operationalizing Machine Learning at Scale
Most machine learning projects never make it to production. Not because the models are bad, but because managing stateful training pipelines and dynamic model drift in production requires platform engineering, not just data science. Here is how I approach MLOps, MLflow, and Kubeflow without the textbook fluff.
Observing the Linux Kernel with eBPF
Running sandboxed programs inside the Linux kernel to trace system calls and inspect packet flow without patching source code or loading unstable kernel modules.