Rubin
R&D Rubin Lab · R&D

The Lab

Rubin Lab is where we do our research. We design the blueprints, the ontology and the onboarding engine that turn a company's documents and tools into a knowledge graph, and we measure every part of it on every release.

  1. 01 Blueprint

    A metamodel for the vertical.

    Before onboarding we draw the blueprint: the metamodel of a vertical. For consumer goods that is accounts, SKUs, contracts, rebate rules, campaigns and the relation types between them.

    Vertical
    Consumer goods
    Classes
    38
    Relation types
    64

  2. 02 Ontology

    Entities, relations, attributes.

    The blueprint becomes the company's ontology. An entity is a thing the business tracks, a relation is how two of them connect, an attribute is a typed fact about one. Each is defined, constrained and layered.

    Entity types
    38
    Relation types
    64
    Attributes
    412

  3. 03 Onboarding

    Every source, read against the blueprint.

    Onboarding scans the company's documents and tools against the blueprint. Each record is matched to a class, duplicates are resolved and every fact keeps the source it came from.

    Sources
    4 systems
    Documents
    8,214
    Records
    1.2M

  4. 04 Knowledge graph

    The company, as one graph.

    What comes out is the knowledge base: entities joined by their relations, versioned over time and kept current as the sources change. Every agent in Rubin answers from it.

    Entities
    12,480
    Relations
    41,902
    Refresh
    live

What we work on

  1. R-01 Vertical blueprints

    Metamodels for an industry, drawn before onboarding: the classes, relation types and attributes a consumer goods or retail company runs on.

  2. R-02 Ontology design

    Turning a blueprint into one company's ontology, with types, constraints and definitions its teams can read and correct.

  3. R-03 Onboarding and extraction

    Scanning documents and tools against the blueprint, matching every record to a class and resolving the same thing spelled differently across systems.

  4. R-04 Graph memory

    Keeping every version of every fact in the knowledge graph, so it can answer what was true, when, and what changed it.

  5. R-05 Evaluation

    Held-out company documents and question sets that measure extraction, resolution and grounding on every change we ship.