Why this matters now
Most AI teaching fragments theory and tooling. This curriculum's core insight is that engineers learn fastest when they build the primitives themselves: derive the math, implement a minimal version, then run the same thing with production libraries. That loop turns shallow API consumption into engineering skill.
What Sets It Apart
- Explicit learning spine: 20 ordered phases (math → agents → production) so later topics never feel like disconnected tricks — you can skip lower phases if already proficient but the syllabus is designed to stack knowledge.
- Artifact-first lessons: each of the 435 lessons produces a reusable deliverable (prompt, SKILL.md, agent, MCP server) you can drop into real workflows instead of checklist exercises.
- Multi-language and production-aware: parallel implementations and production-focused chapters (inference, quantization, MCP, observability, deployment) make it useful for practitioners, not just researchers.
- Toolchain & installability: scripts to scaffold a workbench, install skills (SkillKit), and build a machine-readable catalog reduce friction for teams adopting the curriculum.
Who It's For & Tradeoffs
Great fit if you want a rigorous, engineer-first path to ship real AI systems — learners who prefer coding, math-first explanations, and end-to-end projects (capstones, agent workbenches). Look elsewhere if you need bite-sized video tutorials, purely conceptual surveys, or turnkey SaaS products; this repo expects time investment (hundreds of hours) and familiarity with coding tools.
Where It Fits
Use this as a learning spine for teams building internal AI competency, a university-style course that emphasizes implementable primitives, or as a source of reusable agent skills and production patterns for engineering teams.
