pacoxu/AI-Infra

What it solves

The AI infrastructure stack is complex and evolving quickly. Engineers often struggle to know which components matter, how they relate, and what to learn first. This project provides a curated landscape and structured learning path to demystify that stack, especially in the Kubernetes and cloud-native ecosystem.

How it works

It offers a visual landscape diagram with axes: horizontal maps prototype vs. kernel/runtime maturity, vertical maps infrastructure layer (bottom) to application layer (top). It categorizes key components and provides a learning path organized into sections: kernel & runtime, scheduling & workloads, model inference & runtime optimization, AI gateway & agentic workflow, training on Kubernetes, observability of AI workloads, and ecosystem initiatives. Each section links to detailed documentation covering topics like GPU scheduling, inference engines, parallel training strategies, and agent platforms. The repository also includes a goal achievement chart inspired by baseball player Shohei Ohtani's methodology, helping architects plan their professional and personal growth.

Who it’s for

Engineers and architects who build, operate, or plan to build AI infrastructure, especially those working with Kubernetes and cloud-native technologies. It's useful for both beginners looking for a structured path and experienced practitioners wanting a comprehensive reference.

Highlights

• Curated landscape diagram mapping AI infra components by maturity and layer
• Structured learning path with sections on kernel/runtime, scheduling, inference, agentic workflows, training, and observability
• Extensive documentation covering Kubernetes internals, GPU management, inference optimization, training parallelism, and more
• Lists of relevant open-source projects and community initiatives
• A goal achievement chart for cloud native AI infra architects