Nodes, Edges, and Properties: The Building Blocks

Welcome to the technical foundation of the course. To build a Graph RAG system, you must stop being just an "AI Developer" and become a "Graph Architect." This requires a shared vocabulary. Whether you use Neo4j, Neptune, or a custom Python graph, the fundamental concepts remain the same.

In this lesson, we will explore the three "Building Blocks" of the Property Graph: Nodes (The Actors), Edges (The Actions), and Properties (The Details). We will learn how to transform a sentence into a structured graph element and how to design "Rich Edges" that prevent your graph from becoming a simple, flat list.

---

1. Nodes: The "Nouns" of the Graph

Definition: A Node (or Vertex) represents an object or entity in your world.

• In a Corporate Graph: Employee, Project, Office, Budget.
• In a Scientific Graph: Molecule, Reaction, Paper, Scientist.

Each node should have at least one Label (e.g., :Person). This allows the system to filter nodes quickly. "Show me all nodes labeled :Symptom."

---

2. Edges: The "Verbs" of the Graph

Definition: An Edge (or Relationship) defines the connection between two nodes.

Edges are what create the "Graphiness" of your data. Without edges, you just have a very expensive list.

• Example: (Person) -[:WORKS_AT]-> (Company)

The Directional Rule:

In many graph systems, edges are directed. An edge from A -> B is not the same as B -> A.

• (Sudeep) -[:MANAGES]-> (Jane) means Sudeep is the boss.
• (Jane) -[:MANAGES]-> (Sudeep) means Jane is the boss.

---

3. Properties: The "Adjectives" of the Graph

Definition: Properties are Key-Value pairs attached to nodes or edges. This is what makes a "Property Graph" so much more powerful than a simple mathematical graph.

Node Properties:

Instead of creating a node for "Age," you add a property to the Person node: age: 32.

Edge Properties:

This is a "Pro" move in Graph RAG. You can store metadata on the relationship itself.

• Example: (Sudeep) -[:WORKS_AT {startDate: '2020-01-01', status: 'Full-time'}]-> (London Office)

Why this matters for RAG: Your agent can now answer: "Where did Sudeep work BEFORE 2021?" by filtering the startDate property on the edges.

graph LR
    A((Person: Sudeep)) -- WORKS_AT {role: 'Lead'} --> B((Office: London))
    A -- BORN_ON --> C((Date: 1990))
    
    style A fill:#4285F4,color:#fff
    style B fill:#34A853,color:#fff
    style C fill:#f4b400,color:#fff

---

4. Modeling Exercise: Sentences to Triplets

Let's practice "Graph Thinking" by mapping natural language to our building blocks.

Sentence: "Apollo 11 landed on the Moon in July 1969 with Neil Armstrong as commander."

The Graph Representation:

1. Node: [Apollo 11] (Label: :Mission)
2. Node: [The Moon] (Label: :CelestialBody)
3. Node: [Neil Armstrong] (Label: :Person)
4. Edge: (Apollo 11) -[:LANDED_ON {year: 1969}]-> (The Moon)
5. Edge: (Neil Armstrong) -[:COMMANDED]-> (Apollo 11)

By breaking the sentence down, we have created a Queryable Structure. We can now ask: "Which celestial bodies did COMMANDED missions land on?"

---

5. Implementation: Defining a Property Graph in Python

Using the networkx library, we can build a programmatic Property Graph.

import networkx as nx

# 1. Create the Graph object
G = nx.MultiDiGraph()

# 2. Add Nodes with properties
G.add_node("Sudeep", label="Person", age=32)
G.add_node("London", label="City", population=9000000)

# 3. Add Edges with properties
G.add_edge("Sudeep", "London", type="LIVES_IN", since=2015)

# 4. Query the building blocks
node_data = G.nodes["Sudeep"]
edge_data = G.get_edge_data("Sudeep", "London")[0]

print(f"Node Properties: {node_data}")
print(f"Relationship Property: since {edge_data['since']}")

# RESULT:
# Node Properties: {'label': 'Person', 'age': 32}
# Relationship Property: since 2015

---

6. Summary and Exercises

Nodes, Edges, and Properties are the "Holy Trinity" of Graph RAG.

• Nodes are the entities (Nouns).
• Edges are the relationships (Verbs).
• Properties are the metadata (Adjectives).

Exercises

1. Modeling Challenge: Take a single paragraph from a news article. Identify 5 Nodes and the 5 Edges that connect them.
2. Property vs. Node: Decisions, decisions! Would you model "Salary" as a Node or as a Property on the :WORKS_AT relationship? Hint: Do you need to aggregate "Salary" as a concept, or just know its value for a specific employee?
3. Visual Draft: Draw a graph of your family. Use directed edges for [:PARENT_OF]. Are there any nodes that have multiple incoming edges? This is a "Fan-in" topology.

In the next lesson, we will look at Directed vs Undirected Graphs and how the flow of information changes your agent's reasoning.