The Best KG System Is Useless With One User: Tech vs PMF

I built an extraordinarily complex Knowledge Graph system. Four-layer temporal architecture. Bi-temporal nodes. Semantic search with RRF fusion. Community detection. Ontology validation with 18 relationship types. Contradiction detection. Confidence scoring with monthly decay. Cross-project entity promotion.

Seven registered accounts. One active user — me.

Every line of code is technically correct. Every line is useless because nobody except me sees these graphs.

How This Happens

Engineering satisfaction. Building knowledge graphs presents genuinely stimulating problems — ontology design, temporal reasoning, entity resolution. AI tools now make these achievable in hours instead of months. So you do it. The work feels productive. The graph grows. The metrics are “capabilities implemented.”

Survivorship bias. You read about enterprise success with Neo4j and knowledge graphs. You don’t read about the thousands of projects where the graph was built, maintained for months, and eventually discarded because nobody used it.

Wrong metrics. I measured “38 capabilities shipped.” I should have measured “users who derived value this week.” Those are different numbers. One of them was growing fast.

When Tech-First Actually Makes Sense

There are three legitimate reasons to build infrastructure before finding users:

Enterprise sales cycles. Enterprises buy platforms that can pass security reviews — compliance, audit trails, data sovereignty. Without these features, you never get past the procurement team. The infrastructure is the entry ticket.

Technical moat. A temporal graph with proper edge-case handling is genuinely hard to replicate in a weekend. If the graph is your competitive advantage, building it deeply and correctly makes sense.

Infrastructure as product. When the Knowledge Graph is the thing you’re selling — not a feature supporting something else you’re selling — then building it before PMF is the product development, not premature optimization.

What I’m Doing Now

Froze Knowledge Graph development. Shifted the ratio: 80% distribution (cold outreach, content, partnerships), 20% product.

Extracted the KG demo into separate materials targeting enterprise marketing — that’s the audience where infrastructure-first actually converts.

Redefined success from “shipped capabilities” to “users who saw value this week.”

The Rule

Build infrastructure after finding 10 people willing to pay for it.

Or before searching — if your buyer is enterprise and the infrastructure is the product itself.

Everything else is engineering pleasure at your own expense.