Skip to content

47. Semantic Memory

Mini-Project: Semantic Memory: Knowledge Graph + Vector Store

A semantic memory system combining a knowledge graph for exact entity lookups with a vector store for open-ended conceptual queries, powering a customer success agent that answers questions using both structured and semantic retrieval.

View on GitHub


What It Is

Semantic Memory stores general knowledge, facts, and concepts independent of when they were learned. Unlike episodic memory (specific events), semantic memory stores what the agent knows -- factual relationships, entity definitions, and conceptual understanding. Like knowing "Paris is the capital of France" without remembering when you learned it.

What It Stores (Examples)

  • User profiles and preferences (accumulated over time)
  • Domain knowledge (product specs, policies, technical docs)
  • Entity relationships (customer -> orders -> products)
  • Learned concepts and definitions

How It's Implemented

Combines knowledge graphs (for structured relationships) with vector stores (for unstructured knowledge).

Mermaid Diagram

flowchart TD
    A[Query] --> B{What Type?}
    B -->|Factual| C[Knowledge Graph]
    B -->|Conceptual| D[Vector Store]
    C --> E[Entity + Relations]
    D --> F[Similar Concepts]
    E --> G[Agent Uses Knowledge]
    F --> G

    style C fill:#2196F3,color:#fff
    style D fill:#9C27B0,color:#fff
    style G fill:#FF5722,color:#fff