A growing library of the resources I use to learn AI infrastructure,
plus one flagship project where I turn that knowledge into a working platform.
Learning library · World Model Platform · Progress · Roadmap
Each learning folder represents an independent course, handbook, repository, or reading list. The world-model-platform/ is the only large build and brings the strongest ideas together in one Kubernetes-native system.
AI-Infrastructure/
├── zero-hero-course/ # selected introductory labs
├── stanford-cs336-.../ # language modeling from first principles
├── camilo-inference-recommendations/ # curated inference study path
├── jax-scaling-book/ # student notes for How to Scale Your Model
├── world-model-platform/ # the flagship platform project
├── research/ # researched future learning sources
├── assets/ # diagrams and artwork
├── ROADMAP.md
├── CONTRIBUTING.md
├── SECURITY.md
└── LICENSE
One source gets one top-level folder. New folders appear only when study begins.
An external introduction to cloud, containers, Kubernetes, GPUs, training, serving, observability, MLOps, and security. Roughly 36 useful labs were retained; repetitive or low-value sessions were skipped.
A fundamental systems course covering architecture, resource accounting, GPU optimization, distributed training, scaling, data, evaluation, inference, and post-training. Current focus: Lecture 3, Architectures and Hyperparameters.
A practical path curated by Camilo (@jcfernandezpr), from inference arithmetic and memory fundamentals to vLLM benchmarking, GPU kernels, engines, and research papers.
Open the inference optimization resources
A student-side study path for How to Scale Your Model: rooflines, Transformer math, parallelism, training and inference cost, and GPU/TPU tradeoffs. Includes a local PDF.
Open the scaling-book study folder
A compact Kubernetes-native platform for the full model lifecycle:
collect -> validate -> train -> evaluate -> register -> serve -> observe -> improve
The project will cover reproducible infrastructure, data pipelines, GPU training, experiment tracking, serving, observability, security, cost controls, and runbooks.
This is the target architecture; implementation is still in its foundation phase. Explore the project.
- Library: 4 learning sources indexed
- Active study: Stanford CS336 Lecture 3
- Queued: How to Scale Your Model
- Platform: scope, architecture, and milestones defined
See the learning radar for researched future resources and ROADMAP.md for active work.
Open any learning folder independently, or follow the World Model Platform to see the concepts combined. Requirements vary, so each source documents its own tools, hardware, cost, verification, and cleanup.
Learning folders preserve their original authorship and licensing. The repository-level Apache License 2.0 applies only to my original contributions unless stated otherwise.
Corrections and attribution fixes are welcome. See CONTRIBUTING.md. Security-sensitive reports should follow SECURITY.md.