Aviation Safety - Demo Script¶

This demo showcases Aletheia's GraphRAG capabilities using an aviation safety incident database containing 10 European aviation occurrences from 2024.

Prerequisites¶

1. FalkorDB running: redis-cli ping should return PONG
2. Claude Desktop with MCP servers configured:
3. Graph - Aviation Safety: Direct Cypher queries
4. Aletheia - Aviation Safety: Semantic search via Graphiti

Part 1: Graph Overview (FalkorDB UI)¶

Open FalkorDB Browser UI at http://localhost:3000

1.1 Graph Statistics¶

// Count nodes by type
MATCH (n)
RETURN labels(n) as type, count(*) as count
ORDER BY count DESC

Expected results:

Talking point: The graph contains 10 aviation safety occurrences with their related entities - aircraft, airports, operators, manufacturers, and countries.

1.2 Relationship Types¶

// Count relationships by type
MATCH ()-[r]->()
RETURN type(r) as relationship, count(*) as count
ORDER BY count DESC

Expected results:

Talking point: Rich relationship structure connects incidents to aircraft, operators, locations, and manufacturers.

1.3 Visual: Incident Network¶

// Visualize an incident with all connections
MATCH (o:Occurrence {name: 'Occurrence 2024-0412-EU'})-[r]->(e)
RETURN o, r, e

Talking point: Show the visual graph - one incident connected to aircraft (Embraer ERJ-195), operator (TAP Air Portugal), airport (En route near Barcelona), manufacturer (Embraer).

1.4 Visual: Country-based View¶

// See all incidents by country
MATCH (c:Country)<-[:LOCATED_IN|RELATES_TO]-(ap:Airport)<-[:HAS_AIRPORT]-(o:Occurrence)
RETURN c, ap, o

Talking point: France has the most incidents (3), followed by Spain (3).

Part 2: Semantic Search with Aletheia MCP (Claude Desktop)¶

Switch to Claude Desktop with "Aletheia - Aviation Safety" MCP enabled

Demo Question 1: Semantic Cause Lookup¶

Question:

"What caused the turbulence-related injuries near Barcelona?"

Expected answer:

Unpredicted clear air turbulence associated with jetstream boundary

Capability demonstrated: - Semantic matching: No keyword "Barcelona turbulence" exists - system must understand "near Barcelona" refers to the en-route incident - Entity resolution: Connects location description to specific occurrence - Fact extraction: Retrieves the primary cause from incident details

Why this is impressive: The graph stores the location as "En route, 45 nm west of Barcelona (LEBL)" - semantic search understands "near Barcelona" matches this.

Demo Question 2: Location-based Incident Discovery¶

Question:

"What incident occurred at Nice Cote d'Azur Airport?"

Expected answer:

Wind shear encounter during approach caused by catabatic winds (Occurrence 2024-0912-EU)

Capability demonstrated: - Airport name matching: Handles French accents and variations - Incident summarization: Returns concise description, not raw data - Relationship traversal: Airport -> Occurrence -> Details

Demo Question 3: Entity Description (Narrative Generation)¶

Question:

"Describe incident 2024-0412-EU"

Expected answer:

A TAP Air Portugal Embraer ERJ-195 encountered severe clear air turbulence near Barcelona, injuring two cabin crew members, and diverted to Barcelona El Prat Airport

Capability demonstrated: - Entity aggregation: Combines facts from multiple relationships - Natural language generation: Produces coherent narrative from graph data - Multi-attribute retrieval: Aircraft type, operator, cause, outcome, location

Demo Question 4: Multi-hop Relationship Query¶

Question:

"What incident involved the ITA Airways ATR 72?"

Expected answer:

Display management computer software anomaly during approach to Milan Malpensa (Occurrence 2024-0678-EU)

Capability demonstrated: - 2-hop traversal: Operator -> Aircraft -> Occurrence - Entity disambiguation: ITA Airways operates one specific aircraft in this dataset - Context enrichment: Adds location and incident type to answer

Demo Question 5: Specific Event Type Discovery¶

Question:

"What incident required a MAYDAY declaration during cruise?"

Expected answer:

Incident 2024-0856-EU where smoke smell in the cockpit led to MAYDAY and diversion to Barcelona

Capability demonstrated: - Semantic event matching: "MAYDAY declaration" is embedded in incident narrative - Flight phase filtering: "during cruise" narrows the search - Complex fact retrieval: Matches technical aviation terminology

Part 3: Comparative Analysis (FalkorDB Direct Query)¶

Switch to "Graph - Aviation Safety" MCP or FalkorDB UI

3.1 Show the Structured Query Version¶

For the same question "What incident involved the ITA Airways ATR 72?":

MATCH (op:Operator {name: 'ITA Airways'})-[:HAS_OPERATOR]-(o:Occurrence)-[:HAS_AIRCRAFT]->(a:Aircraft)
WHERE a.name CONTAINS 'ATR'
RETURN o.name, a.name, o.summary

Talking point: Cypher requires knowing: 1. Exact operator name ("ITA Airways") 2. Relationship direction and type 3. Aircraft naming convention

Aletheia handles variations like "ITA", "ITA Airways", "ATR 72", "ATR 72-600" automatically.

3.2 Graph Traversal for Relationships¶

// Find all incidents related to Air France group (Air France + HOP!)
MATCH (op:Operator)-[:HAS_OPERATOR]-(o:Occurrence)
WHERE op.name CONTAINS 'France'
RETURN op.name, o.name, o.summary

Expected results:

Part 4: Advanced Queries (FalkorDB)¶

4.1 Manufacturer Analysis¶

// Count incidents by aircraft manufacturer
MATCH (m:Manufacturer)<-[:MANUFACTURED_BY]-(a:Aircraft)<-[:HAS_AIRCRAFT]-(o:Occurrence)
RETURN m.name as manufacturer, count(o) as incidents
ORDER BY incidents DESC

4.2 Country Distribution¶

// Incidents by country
MATCH (c:Country)<-[:RELATES_TO]-(o:Occurrence)
RETURN c.name, count(o) as incidents
ORDER BY incidents DESC

4.3 Full Incident Path¶

// Trace complete incident context
MATCH path = (m:Manufacturer)<-[:MANUFACTURED_BY]-(a:Aircraft)<-[:HAS_AIRCRAFT]-(o:Occurrence)-[:HAS_AIRPORT]->(ap:Airport)-[:LOCATED_IN]->(c:Country)
WHERE o.name = 'Occurrence 2024-0289-EU'
RETURN path

Talking point: Show the full relationship chain: Boeing -> 737-800 -> Incident -> Frankfurt -> Germany

Key Takeaways¶

1. Semantic Understanding: Aletheia understands natural language questions without requiring exact terminology or Cypher syntax

2. Relationship Traversal: Automatically follows graph relationships to gather comprehensive answers

3. Context Enrichment: Combines information from multiple nodes to provide rich, contextual responses

4. Flexible Matching: Handles variations in entity names, abbreviations, and descriptions

5. Dual Access:

6. Aletheia MCP: Natural language for analysts and investigators
7. Graph MCP: Direct Cypher for developers and precise queries

Appendix: Entity Reference¶

Demo Flow Summary¶