AI Course Program
A curriculum designer builds an AI specialization course. The ontology mixes topics, skills, concepts, and tools — all in one graph. The course program is a profile; coverage checks which topics have teaching materials at sufficient depth.
01 Ontology — 4 kinds, 18 nodes in the graph
A course ontology is not just skills — it includes topic nodes for curriculum structure, concept nodes for theory, and product nodes for tools. All in one graph, all linked.
LLMs & NLP
Transformer architecture, prompting, fine-tuning, RAG, evaluation. Central module of the course.
Prompt Engineering
Chain-of-thought, few-shot, structured output, system prompts, prompt versioning.
Transformer
Self-attention, multi-head attention, positional encoding, encoder-decoder architecture.
LangChain
LLM orchestration framework: chains, agents, retrievers, memory, tool use.
id: nlp-llms
name: LLMs & NLP
kind: topic
description: Transformer architecture, prompting, fine-tuning, RAG…
depends_on: [ai-fundamentals, machine-learning]
links_to: [skill:prompt-engineering, concept:transformer, product:langchain]
02 Course Profile — ai-course · program · threshold 0.7
The course program selects topics and skills from the ontology. Each topic has a required depth (how thoroughly it's taught) and a weight in the overall course. Weights sum to 1.
| Topic | Kind | Depth | Weight |
|---|---|---|---|
| nlp-llms | topic | 0.9 | 0.25 |
| ai-fundamentals | topic | 0.8 | 0.20 |
| machine-learning | topic | 0.7 | 0.15 |
| deep-learning | topic | 0.7 | 0.15 |
| mlops | topic | 0.6 | 0.10 |
| ai-ethics | topic | 0.5 | 0.05 |
| computer-vision | topic | 0.4 | 0.05 |
| prompt-engineering | skill | 0.8 | — |
| rag-deployment | skill | 0.7 | — |
| llm-fine-tuning | skill | 0.6 | — |
profile: ai-course level: program threshold: 0.7
requirements:
- node: nlp-llms depth: 0.9 weight: 0.25
- node: ai-fundamentals depth: 0.8 weight: 0.20
- … 5 more topics + 3 skills …
03 Coverage — lecture materials, exercises, labs checked per topic
Each topic is checked against the question/assignment library. Same fixed gap definition — same ontology + same profile → same gaps in every implementation.
| Topic | Materials | Depth | Status | |
|---|---|---|---|---|
| nlp-llms | 12 | 0.9 | ok | |
| ai-fundamentals | 8 | 0.8 | ok | |
| machine-learning | 6 | 0.7 | ok | |
| deep-learning | 5 | 0.7 | ok | |
| mlops | 4 | 0.50 → need 0.6 | thin | |
| ai-ethics | 3 | 0.35 → need 0.5 | thin | |
| computer-vision | 4 | 0.40 → need 0.4 | ok |
→ Add materials with difficulty ≥ 0.50 for mlops and ≥ 0.35 for ai-ethics.
04 Export — course-ai-program.json
Everything derived is generated: a course package with lesson plans, assessment questions, and reading lists — no hand-written syllabi.
schemaVersion: "sdm.export.test/v1",
profile: "ai-course", level: "program", threshold: 0.7,
topics: [ /* 7 topics · depth · weight */ ],
materials: [ /* 42 items · lectures, labs, quizzes */ ],
glossary: [ /* 15 terms · transformer, RAG, LoRA… */ ]
}