Varun Rajput·Nov 9Why May a Model Get Slower on a Bigger Machine?When two libraries both assume they own every CPU core, adding cores can make things slower. Here’s why, and how to spot it.A response icon1A response icon1
Varun Rajput·Oct 29How to accurately time training and inference stepsLet’s suppose you’re training an image classifier on a GPU. Every training step does two things:
Varun Rajput·Jul 14GPU Sharing in Production Kubernetes: When MIG and MPS Actually Matter and When Time-Slicing Is…Because giving a 14GB model a full 80GB A100 and calling it “allocated” is a problem worth solving
Varun Rajput·Jul 8Feature Engineering PipelineThe ML engineer owns the math. The platform engineer owns every point where that math touches infrastructure.A response icon1A response icon1
Varun Rajput·Jul 4LLM Benchmarking for Internal Hosting: How to Pick the Right ModelThe model selection and cost-quality analysis that MLOps engineers actually doA response icon2A response icon2
Varun Rajput·Jun 2The Infrastructure Side of RAG That Most Tutorials SkipWhat actually runs behind the scenes when a user sends a query to a production RAG system
Varun Rajput·May 21Things I Learned Building an End-to-End ML Pipeline on Kubernetes: From Validated Data to Live…Part 2 of an MLOps End-to-End series — 60 models, fully automated, one Airflow DAGA response icon1A response icon1
Varun Rajput·May 8System Design Interview: How would I Solve a Real ETL Pipeline Question Step by StepThe approach that matters more than the answer: clarifying questions, tradeoffs, and the mistakes most candidates make
Varun Rajput·Apr 29MLOps End to End: Part 1 ML Foundation — Real ETL Pipeline on KubernetesWhy your ML project should start with data infrastructure, not a Jupyter notebook
Varun Rajput·Apr 13“CI/CD for MLOps” is not what I thoughtAlmost every MLOps job description mentions CI/CD. Sounds easy if you come from DevOps. You’ve built Jenkins pipelines, GitHub Actions, and…