Wikipedia-style expert knowledge base structured for LLMs
Problem
Power users of Claude and other assistants find the models lack deep, structured expert knowledge — answers are generic because there is no curated, expert-written knowledge graph the assistant can draw from. Existing wikis are prose-heavy and not structured for machine consumption.
Opportunity
A curated 'expert Wikipedia' with content structured as a knowledge graph, consumable by LLMs via MCP/API, where domain experts contribute and the assistant cites them — a trust and depth layer for AI answers.
Market analysis
Demand for verified, structured knowledge feeding LLMs is real, but it is being met by enterprise data licensing, not community wikis: Golden's venture-backed knowledge graph ended up absorbed into ComplyAdvantage, while getAbstract and Turing sell curated content to AI buyers directly.
Market · AI labs and assistant power users; demand exists but is concentrated in B2B data licensing, not consumer subscription.
Pricing · Verified-knowledge players monetize via enterprise licensing (getAbstract for enterprise AI; Turing-style expert data pipelines for labs) — not a model a solo builder can enter.
Pros
- + Structuring knowledge for machine consumption via MCP is a timely angle.
- + Expert-verified citations address a documented weakness of LLM answers.
- + A single vertical domain (one MCP server) would be a feasible starting scope.
Cons
- − Community expert contribution at scale is a decades-long cold-start problem.
- − Golden, with a16z backing, is a cautionary tale: absorbed into a compliance company.
- − The buyers who pay (AI labs) already source expert data through B2B pipelines.
- − Publishers and encyclopedias are adding LLM-facing APIs themselves.
Source
Hacker News (Ask HN)
The cautionary tale here is Golden: a self-constructing, expert-mapped knowledge graph with a16z money behind it, and the ending was an acquihire into a compliance data company — the graph turned out to be worth more as enterprise data than as a public wiki. That points at the structural problem: the parties who genuinely pay for verified expert knowledge are AI labs and enterprises, and they buy it as licensed data pipelines, not as a community product. For a solo builder the only sane scope is one narrow domain shipped as an MCP server that assistants can cite — essentially becoming a trusted micro-publisher — and even then the moat is reputational rather than technical.