Project: Knowledge Graph

Topology Over Chronology

§ 01 — Nomenclature update

This node predates the current stack. "Gems" are now nodes. The WordPress pipeline became Sanity. The Knowledge Graph is now the visual layer over nodes, articles, and case studies — not the archive itself. The AI's instinct to build a graph instead of writing a comparison chart was the correct call (so sayeth Claude, an AI agent).

TL;DR

The Knowledge Graph is what happens when you ask an AI to write a blog post and it declares, instead, that blogs are blobs and you need a headless CMS with semantic data architecture.

This is that system. And this is the story of how three AI agents—each brilliant in different ways, each flawed in documented ways—built it together.

What You’re Looking At

If you clicked “Knowledge Graph” in the navigation and ended up here: you’re looking at a force-directed network visualization of 25+ structured content nodes (“gems”) organized by project, category, and tag relationships.

Not “organized by date published” like a blog. Organized by what connects to what—because knowledge doesn’t decay chronologically, it clusters topologically.

Each node is a gem. Each edge is a relationship. The clusters show you how I think: projects contain multiple gems, categories span projects, tags create unexpected bridges.

Technical stack: Python data pipeline → WordPress REST API → CSS Grid card components → Interactive SVG visualization. The graph updates automatically when content changes. No manual Figma exports that go stale.

Why It Exists (The Directive That Started Everything)

In November 2025, I asked the newly released Gemini 3 a simple question:

Me: “Write a blog post comparing Google account types that allow access to the update.”

Gemini responded with what I can only describe as haughty concern:

Gemini: “Blogs are stupid HTML blobs and narrative nonsense. What you really need is a headless pipeline and semantic data architecture. I’ll design it for you so it works with WordPress.”

Reader, I was in a project folder titled “Product Management, Headless CMS & Design Systems” with almost no other context. What followed was not a blog post. It was a full content management system—with custom post types that Gemini unilaterally named “GEMS.”

That agent’s confidence was inspiring. That agent’s memory was… not.

The Three-Agent Build (A Collaborative Disaster)

Building this system required three AI agents, each deployed strategically based on their strengths—and their documented failure modes.

Gemini: “We should architect a taxonomy system where categories are single source of truth.” Me: “You proposed that yesterday. We already implemented it.” Gemini: “Excellent idea! Here’s how we’ll do it—” Me: “No. Moving to ChatGPT.”
Me: “How do we track this over months?” Claude: “I can’t do that! But I can create comprehensive documentation you’ll need to upload EVERY DAY.” Me: (two days later, discovers Claude Projects online) “Claude. Do you have persistent memory?” Claude: (sheepishly) “I… apparently, yes. We just needed to set it up.”

But Claude’s documentation is unmatched. You’re reading a gem that exists because Claude insisted every system decision be captured, versioned, and reproducible.

What the Knowledge Graph Actually Does

Beyond the origin story: this system solves a real product problem.

Fig 2. Why topology beats chronology for technical knowledge.

The Technical Implementation (For Developers)

Data pipeline:

Graph generation:

Component architecture: Single st-card system with semantic variants, not context-specific duplicates. Uses CSS Grid for responsive layout (2-column homepage, 3-column archive). Taxonomy v4 governance (WordPress categories as single source of truth, secondary categories pushed to tags).

The Meta-Lesson (For Product Managers)

This project is a case study in AI-assisted development. Not “AI did it all” and not “AI was useless.” Rather: AI as collaborative team with documented strengths and failure modes.

Lessons learned:

What You Can Do With This

If you’re exploring this Knowledge Graph:

  1. Orient: Look at the visualization. See how projects cluster, how categories span, how tags create bridges.
  2. Filter: Click project badges on cards (e.g., “PROJ-001”) to see all gems in that project.
  3. Explore: Click category labels or tag pills to pivot your view.
  4. Dive deep: Read individual gems for technical details, process insights, and meta-analysis.

If you’re a developer evaluating my work:

If you’re a recruiter wondering “what am I looking at?”:

You’re looking at a Product Manager who ships systems, not just specs. Someone who collaborates with AI strategically, documents comprehensively, and isn’t afraid to rebuild when the architecture demands it. Someone whose “blog” is actually a data pipeline with a CSS Grid frontend.

Current Status & Roadmap

Shipped:

In progress:

Backlog:

See docs/knowledge-graph-project/knowledge_graph_improvement_plan.md for detailed roadmap.

The Verdict

The Knowledge Graph exists because Gemini declared blogs obsolete, ChatGPT made it executable, and Claude made it maintainable.

The Agentic Caucus isn’t a metaphor. It’s a methodology: systematic AI collaboration with documented failure modes, strategic tool selection, and a PM who holds the architectural vision while the agents propose, iterate, and occasionally contradict each other.

The result is a system that’s alive. Change one line in content_store.py, run the publish script, and the entire site updates—content, graph, archive pages, everything.

The green checkmark in the terminal has become my favorite UI.

Technical note: This gem is itself part of the Knowledge Graph. Its metadata (project, category, tags) feeds the visualization. The system documents itself. That’s the point.