Notes and deep dives on AI infrastructure, LLM deployment, MLOps, and building reliable production ML systems.
3 articles
A practical walkthrough of the infrastructure decisions behind serving large language models reliably — from GPU selection to batching, autoscaling, and observability.
Quantization, batching, and right-sizing strategies that reduced our inference bill by 60% while keeping p99 latency flat.
The core building blocks of a production MLOps platform — model registry, CI/CD for models, and safe rollouts with canaries and shadow deployments.