Architecture
How Facets decomposes and answers systems questions
Every query flows through 5 stages: classification through 8 structural dimensions, dimension-aware routing, dual retrieval from a knowledge graph and vector index, LLM synthesis, and grounded response with source evidence.
Built on: George Mobus, Systems Science (Springer, 2022) · LlamaIndex + BGE embeddings · Architecture docs
1 Your Question
Any natural-language question — about a real-world system, a concept, or a hypothetical scenario.
Expanded with systems science synonyms to improve retrieval
2 8-Tuple Dimension Classifier
C
Components
What constitutes the system
N
Network
How parts connect internally
E
Environment
What lies outside the boundary
G
External Flows
What crosses the boundary
B
Boundary
Where the system ends
T
Transforms
How inputs become outputs
H
History
What the system remembers
Dt
Time Scale
How fast or slow it operates
Intensity: full · light · none
Which dimensions are active determines which sources are searched
C
N
E
G
B
→
Concept Relationships
T
H
Dt
→
Textbook Passages
Structure dimensions (C, N, E, G, B) search the concept graph. Process dimensions (T, H, Dt) search textbook passages.
3 Retrieval
Concept Relationships (knowledge graph)
Follows relationships between concepts, e.g. “boundary defines system”
Used for: Components, Network, Environment, Flows, Boundary
734 mapped relationships
LlamaIndex KnowledgeGraphIndex
Textbook Passages (vector index)
Searches Mobus's Systems Science across 16 chapters by meaning
Used for: Transforms, History, Time Scale
554 indexed passages
LlamaIndex VectorStoreIndex · BGE-small
4 Synthesis
LLM Synthesis
Weighs the retrieved sources and composes a grounded answer through a systems science lens
Claude Haiku (cloud)
Ollama (local)
5 Response
TL;DR bullets + full analysis, annotated with active dimensions, retrieval route, confidence score, and source evidence from the textbook.