Knowledge that knows how it connects

Nodes and typed relationships

A knowledge graph is a way of storing knowledge as a network rather than as a stack of documents. It has two building blocks. A node is a single thing worth talking about: a person, a book, a verse, a concept, a research paper. A relationship (also called an edge) is a labeled connection between two nodes — and the label is the important part. The connection is not just "these two things are related somehow"; it is typed, meaning it states exactly how they relate: this teacher taught that student, this teaching cites that verse, this paper discusses that concept.

Because every connection carries a type, a graph can answer questions that would otherwise require a human to read everything. "Which teachings cite this verse?" is a single hop along citation edges. "Which sages connect these two schools of thought?" is a path search across teacher–student edges. The structure itself does work that plain text cannot.

Provenance: every claim carries its source

Provenance means the recorded origin of a piece of information — where it came from and how you can check it. In a well-built knowledge graph, provenance is not an afterthought; it is attached to the data itself. A node representing a teaching does not merely store the teaching's text: it stores the citation pointing at the original source, so any claim in the graph can be traced back and verified.

This is the discipline that separates a knowledge graph from a loose web of assertions. A graph without provenance is just a diagram of opinions. A graph with provenance is a checkable map of what the sources actually say.

Why a graph beats a pile of documents

Documents are how knowledge is written; they are not how knowledge is shaped. Real knowledge is connected: a commentary refers to a verse, which appears in a weekly reading, which a later sage discusses, who studied under an earlier sage. In flat documents, each of those connections exists only implicitly — as a reference a human reader must notice, look up, and hold in memory.

A graph makes those connections explicit and machine-traversable, which changes what is possible:

Two working graphs you can trace

None of the above is hypothetical. This network runs two knowledge graphs today, both documented publicly on the jew.tech hub.

The first is the Torah knowledge graph: 157,143 nodes joined by 215,426 relationships, covering 33,323 verses and 17,147 cited teachings — every one of those teachings stored with its Hebrew text, its citation, and an embedding (a numeric representation that lets software find similar passages). Its 15,910 teaching-to-verse citation edges are provenance made concrete: each teaching points at the exact verses it draws on. It also maps the human chain of transmission, with 2,504 rabbi-to-rabbi edges joining 2,185 sages across the generations.

The second comes from research literature: a corpus of 22,841 research papers assembled into a 202,816-node concept graph, where papers and the concepts they discuss are the nodes and the edges record which paper engages which idea. The same principles — typed relationships plus provenance — applied to a completely different body of knowledge.

Part of a larger project

This site is one node in a network of related projects exploring connected knowledge:

The hub for the whole network — the graphs, the data, and the discoveries — is jew.tech.