A 12-week semantic web curriculum, worked in public.
RDF, OWL, SPARQL, and modern knowledge graphs — from first principles to a deployed hybrid LLM + knowledge graph application. Every artifact, note, and stumbling point published openly.
Why publish
learning openly?
Most learning artifacts describe finished states. This one shows the process. The redos, the failed exercises, and the wrong turns are part of the published value, not embarrassments to hide.
Forcing function
Public commitments stall less. Knowing the work is visible turns “I should write notes” into “I have to ship notes.”
Compounding portfolio
Each module produces real artifacts: published ontologies, evaluation notebooks, deployed demos. The curriculum is the portfolio.
Honest signal
A real evaluation, including where graph augmentation hurts retrieval, is more useful than a polished case study that only shows the wins.
Four modules across twelve weeks.
The curriculum builds from foundations (reading and writing the basic data model) to deployment (shipping a system that combines vector and graph retrieval). Module 3 is genuinely the hardest segment — planning slower pacing there matters.
The arc is reminiscent of marathon prep: base, build, peak, taper-and-race. Module 3 is the segment where most of the material is genuinely new ground for me, personally.
Foundations
RDF, Turtle, basic SPARQL. The conceptual core: knowledge as triples with global identifiers.
Modeling
OWL ontology design. The shift from RDF as data format to ontology as design discipline.
Reasoning
Inference, reification, vocabulary alignment, SHACL validation. The conceptually hardest segment.
Shipping
SPARQL UPDATE, deployment, hybrid LLM + knowledge graph capstone.
The semantic web stack, layered.
Each layer builds on the one below it. The curriculum walks up the stack in order: triples first, schemas next, ontologies and reasoning after that. SPARQL is the query language that runs across all layers.
Two existing projects as canvases.
The curriculum doesn’t use toy examples. It uses two real projects that already have substantial work behind them. Each anchors specific parts of the curriculum where the technology shines for it.
Naruto Network Graph
A character network from anime subtitle data — 87 characters across three story arcs, with hand-coded canonical relationships layered over co-appearance edges. The rich categorical structure of the Naruto universe (ninja ranks, villages, jutsu, contested fan canon) makes it an unusually good semantic-web testbed. Pop-culture domains are standard in ontology pedagogy: the Pizza ontology is the canonical Protégé tutorial for the same reason.
Resume Graph Explorer
A career graph with ESCO/SKOS integration — existing Neo4j implementation with skill taxonomy linked to the European Skills, Competences, Qualifications and Occupations vocabulary. The reason it anchors Module 1: ESCO is already a SKOS vocabulary, which means the case for RDF over a labeled property graph is exceptionally sharp here. Returns in Module 3 as a venue for skill inference work.
Each project enters where it adds the most learning value. Resume Graph for the LPG-vs-RDF case where ESCO makes RDF shine. Naruto for the rich modeling and reasoning work. TwinKit as the capstone framework where everything composes.
Everything lives here.
The GitHub repository is the canonical record. Syllabus, weekly progress, exercises, ontologies, notes, and published artifacts all live in one place. Visitors land on the README; followers track PROGRESS.md for weekly updates.
Four platforms, distinct roles.
The work lives across multiple surfaces. Each has a different role and a different voice. GitHub is the canonical record. LinkedIn broadcasts progress. The two blogs handle longer-form thinking that doesn’t belong in either.
The official version. All commits dated. Watch the repo for new artifacts as they land.
Monday progress + one weekly insight. Friday artifact or open question. Hooks first, white space heavy.
Long-form essays on the design decisions, the “what I changed my mind about,” the cog-sci-flavored reflections.
The most personal layer. Working-notebook explorations that don’t fit anywhere else.
This weekend: pre-week setup.
The pre-week is roughly 2–3 hours of setup work spread across morning and afternoon. Front-loading these means Module 1’s first day isn’t lost to environment-fighting.
Morning ~1.5 hours
- Push the initial repo files to GitHub. README, SYLLABUS, PROGRESS, CLAUDE.md, Module 1 README. The repo should look maintained from day one.
- Install Apache Jena + Fuseki locally. Verify it runs on
localhost:3030and I can create a new dataset. - Install Protégé and open it once. Don’t do anything with it yet.
- Bookmark the W3C primer set. RDF 1.1, Turtle, SPARQL 1.1 Query — these are my spec references.
Afternoon ~1 hour
- Pick the LinkedIn launch post variant and schedule for Tuesday or Wednesday mid-morning Central time. Repo link goes in the first comment, not the post body.
- Block 12 Tuesday lunch slots as “synthesis writing” (30 minutes each). That single ritual carries the curriculum forward.
- Order or borrow Allemang, Hendler & Gandon (3rd ed., 2020) if not already on the shelf.
- Skim Hogan et al. introduction (kgbook.org sections 1–2). Vocabulary calibration only — don’t try to read the whole thing.
This week (Module 1 begins) ~6 hours
- Read Allemang chapters 1–3 (RDF foundations, RDFS).
- Exercise 1.1: Wikidata orientation. Three queries through query.wikidata.org. Notes go in
modules/01-foundations/notes/week-01.md. - Exercise 1.2: Hand-write a FOAF graph of myself and three of my projects.
- Begin Exercise 1.3: Resume Graph RDF slice. The Module 1 primary project.
- Write Week 1 synthesis notes. One paragraph, in my own words, on what it felt like to cover the material.
- Post first Monday LinkedIn update. “Week 1 of 12” momentum check.
Placeholders for richer content.
These slots will fill in as each module produces its artifacts. Listed here so the structure is visible from day one and the work has clear targets.
SPARQL playground for ESCO queries
Embedded query editor with sample queries from the Resume Graph Explorer comparison. Run live against a public endpoint.
Naruto ontology in WebVOWL
Interactive class hierarchy explorer rendered from the published Turtle. Click any node to see properties and instances.
Designing an ontology in 3 hours
Time-lapse screen recording of the actual modeling work. The wrong turns visible. Less polish than tutorial videos; more honesty.
Reification approaches, compared
Side-by-side: same fact expressed in classical RDF reification, n-ary, named graphs, RDF-star. Toggle between styles.
Naruto Knowledge Graph Explorer
Public web app. Ask natural-language questions, see generated SPARQL, explore character networks with ontology overlay.
Live progress dashboard
Auto-generated from PROGRESS.md commit history. Modules completed, artifacts shipped, current focus. Embedded on homepage.
The canonical references.
None of the conceptual content in the curriculum is original; the contribution is the path through it. These are the books and specs the curriculum points to repeatedly. The full bookmark set lives in resources/reading-list.md.
One practitioner, working openly.
This curriculum is published under Sensemaking AI, an independent AI and machine learning consulting practice. It’s run solo by Barbara Hidalgo-Sotelo — cognitive scientist, AI/ML consultant, Toastmaster, marathon runner, beekeeper, and longtime builder of quirky-but-rigorous data projects.
Two web homes, two different jobs. barbhs.com is the technical portfolio — projects, experiments, the body of work. sensemaking-ai.com is the practice — applied AI work for people and organizations who want to think clearly with AI, not just move faster because of it.
The curriculum sits between them. It’s how the technical body of work gets developed in the open, and where the practice’s intellectual commitments get tested against real material.
Questions, corrections, and discussion welcome via the repository’s issues. Pull requests on the curriculum itself aren’t accepted — this is a personal learning project — but disagreements about design choices are always interesting.
Exploring messy data, intelligent systems, and what it means to make meaning — through the lens of a cognitive scientist who builds things people actually use.