Part 2: Knowledge Graphs 101¶

Back to the drug interaction problem. You've decided to build a knowledge graph instead of just embedding documents.

You start with aspirin. What do you know about it?

• Aspirin is a drug
• Aspirin is an NSAID
• Aspirin is an anticoagulant
• Aspirin treats pain
• Aspirin treats inflammation
• Aspirin increases bleeding risk

Each of these is a relationship between aspirin and some other concept. Let's write them as triples:

(Aspirin) --IS_A--> (NSAID)
(Aspirin) --IS_A--> (Anticoagulant)
(Aspirin) --TREATS--> (Pain)
(Aspirin) --TREATS--> (Inflammation)
(Aspirin) --INCREASES_RISK--> (Bleeding)

Now warfarin:

(Warfarin) --IS_A--> (Anticoagulant)
(Warfarin) --TREATS--> (Blood Clots)
(Warfarin) --INCREASES_RISK--> (Bleeding)

Look at that. Aspirin and warfarin both connect to "Anticoagulant" and both connect to "Bleeding." The dangerous interaction is visible in the structure.

Concepts and edges¶

A concept (node) is a thing: a drug, a condition, a property, a procedure.

An edge connects two concepts with a semantic relationship.

The relationship type matters. "Aspirin TREATS Pain" is different from "Aspirin CAUSES Pain." The verb carries meaning.

Semantic types¶

qortex uses 10 relationship types:

Type	Meaning	Example
REQUIRES	A needs B	Circuit Breaker REQUIRES Timeout
CONTRADICTS	A and B conflict	Retry CONTRADICTS Fail-Fast
REFINES	A is a specific form of B	Ibuprofen REFINES NSAID
IMPLEMENTS	A is a concrete form of B	Redis IMPLEMENTS Cache
PART_OF	A is a component of B	Handler PART_OF Middleware
USES	A depends on B	Service USES Database
SIMILAR_TO	A and B are analogous	Aspirin SIMILAR_TO Ibuprofen
ALTERNATIVE_TO	A can substitute for B	Tylenol ALTERNATIVE_TO Aspirin
SUPPORTS	A provides evidence for B	Study SUPPORTS Efficacy
CHALLENGES	A provides counter-evidence	Side Effect CHALLENGES Safety

Rich typing is what enables automated rule derivation later. If you know the relationship type, you can generate rules mechanically.

Domains¶

Concepts belong to domains, isolated subgraphs like schemas in a database.

You might have: - A "pharmacology" domain with drug concepts - An "error_handling" domain with software patterns - A "legal" domain with compliance rules

Domains don't interact by default. This prevents cross-contamination when you ingest multiple sources.

Your first knowledge graph¶

from qortex.core.models import ConceptNode, ConceptEdge, RelationType
from qortex.core.memory import InMemoryBackend

backend = InMemoryBackend()
backend.connect()
backend.create_domain("pharmacology", "Drug interactions and effects")

# Add concepts
backend.add_node(ConceptNode(
    id="aspirin",
    name="Aspirin",
    description="NSAID with anticoagulant properties",
    domain="pharmacology",
    source_id="drug_database",
))

backend.add_node(ConceptNode(
    id="anticoagulant",
    name="Anticoagulant",
    description="Substance that prevents blood clotting",
    domain="pharmacology",
    source_id="drug_database",
))

# Add relationship
backend.add_edge(ConceptEdge(
    source_id="aspirin",
    target_id="anticoagulant",
    relation_type=RelationType.REFINES,  # Aspirin is a type of anticoagulant
))

You now have structure. Concepts connected by typed relationships.

But a graph sitting in memory doesn't help anyone. You need to extract actionable rules from it.

What you learned¶

• Knowledge graphs store concepts (nodes) and relationships (edges)
• Edge types carry semantic meaning (REQUIRES vs CONTRADICTS)
• 10 relationship types enable automated rule derivation
• Domains isolate subgraphs to prevent cross-contamination

Next¶

Part 3: The Projection Pipeline: How do you turn a graph into actionable rules?