cogniworks
Knowledge graph

Ontology first.
Knowledge
that reasons.

Prism turns organizational knowledge into a typed, traversable graph. LLMs navigate it in milliseconds and return correct output — not because they got smarter, but because they finally have the knowledge they need.

Prism ingests Forge output directly — documents processed through Forge pipelines become typed nodes and edges in the graph, not summaries.

ontology-firstlegal corpusAI reasoning10ms traversal
prism · insurance KG● live

Query

What are subrogation rights if the
third party is uninsured in a
cross-emirate incident?

3 nodes · 10ms

Prism

Based on UAE Civil Transactions Law and Motor Vehicle Insurance regulations:

→ Subrogation applies under Art. 1034

→ Cross-emirate: CBUAE rules apply

→ Uninsured: Motor Vehicles Act §146

T

Typed

Every node has a kind. Every edge has a predicate. The graph knows what it stores and how pieces relate — making traversal semantic, not statistical.

T

Traversable

"What does this module transitively depend on?" is a single graph traversal. No application code. No BFS loops. The graph handles transitive closure natively.

C

Compounding

Every document added becomes nodes and edges. An invariants doc: eleven new nodes. A claims guide: two thousand sections. The graph grows. It never shrinks.

The problem

AI is intelligent but ignorant.

Architecture decisions in design docs. Domain rules in knowledge bases. Regulatory constraints in compliance guides. Entity relationships in the heads of people who may not be here next year.

No amount of model intelligence compensates for missing knowledge.
A genius with no textbook still fails the exam.

Why not RAG

"What does this module depend on?"

Graph traversal: follow DEPENDS_ON edges to arbitrary depth. RAG finds the most similar paragraph.

"Is this claim covered under this policy?"

Eight-step determination sequence across multiple knowledge nodes. RAG returns one paragraph.

"What invariants govern the service layer?"

Follow GOVERNED_BY edges. The graph knows these explicitly. A vector store can only guess.

The proof

Indian health and motor insurance. The hardest domain we could find.

763 domain entities across nine modules. 31,595 lines of claims processing knowledge. IRDAI regulatory mandates, Motor Vehicles Act constraints, eight-step coverage determination sequences, four-level adjudication authority matrices.

3,239
knowledge nodes
8,488
typed edges
10ms
traversal time
0
hallucinations
Request demo access →

The compounding effect

Day 1

Eleven build invariants encoded. Every LLM interaction from this point forward respects them. No new developer violates them unknowingly.

Day 30

Nine module designs, dependencies, integration contracts. LLM scaffolds new modules that correctly integrate with existing ones.

Day 90

Domain rules encoded. LLM generates domain-correct business logic, not just structurally correct code.

Day 180

3,000+ nodes. The LLM inherits six months of accumulated organizational intelligence — instantly, completely, precisely.