AI Periodic Table Analysis

Mapping Engram’s architecture to Martin Keen’s AI Periodic Table and proposing the missing Gk (Graph Knowledge) element.

The AI Periodic Table

Martin Keen (IBM Master Inventor) released a framework organizing AI technologies into a structured periodic table format. The table uses two dimensions:

  • Rows (Maturity): Primitives → Compositions → Deployment → Emerging
  • Columns (Capability): Reactive → Retrieval → Orchestration → Validation → Models

Martin Keen's AI Periodic Table

The Elements

Row C1 Reactive C2 Retrieval C3 Orchestration C4 Validation C5 Models
R1 Primitives Pr (Prompts) Em (Embeddings) — — Lg (LLM)
R2 Compositions Fc (Function Call) Vx (Vector) Rg (RAG) Gr (Guardrails) Mm (Multimodal)
R3 Deployment Ag (Agent) Ft (Finetune) Fw (Framework) Rt (Red-team) Sm (Small)
R4 Emerging Ma (Multi-agent) Sy (Synthetic) ❓ In (Interpret) Th (Thinking)

The Missing Element

The cell at Row 4 (Emerging) × Column 3 (Orchestration) is empty.

Proposal: Gk (Graph Knowledge)

We propose Gk as the missing element representing Graph-based Knowledge Orchestration.

Definition

Gk (Graph Knowledge): The use of temporal knowledge graphs to dynamically orchestrate agent reasoning, memory retrieval, and context assembly based on semantic relationships rather than static configurations.

Rationale

Aspect Static Frameworks (Fw) Graph Knowledge (Gk)
Routing Logic Predefined chains/DAGs Dynamic, semantic routing
Context Assembly Manual prompt engineering Automatic via graph traversal
Memory Stateless or session-scoped Temporal, multi-session knowledge
Discovery Explicit tool registration Emergent via entity relationships

Why Emerging?

Graph Knowledge is positioned in the Emerging row because:

  1. Nascent tooling: Zep, LangGraph, and similar tools are still maturing
  2. Requires infrastructure: Needs vector DB + graph DB + temporal layer
  3. Paradigm shift: Moves from “prompt engineering” to “context engineering”

Engram Architecture Mapping

Engram implements elements across all five columns and all four rows:

Element Engram Component Implementation
Pr (Prompts) System prompts Agent personalities (Elena, Sage, Marcus)
Em (Embeddings) Zep memory Semantic search across episodes
Lg (LLM) GeminiClient, ClaudeClient Model abstraction layer
Fc (Function Call) MCP Tools 40+ tool definitions
Vx (Vector) Zep vector store Episode and fact retrieval
Rg (RAG) Tri-Search Memory → LLM augmentation
Gr (Guardrails) Auth middleware Enterprise security boundaries
Mm (Multimodal) Imagen 3.0, VoiceLive Image generation, voice I/O
Ag (Agent) Elena, Marcus, Sage Autonomous reasoning agents
Ft (Finetune) — Not yet implemented
Fw (Framework) Temporal workflows Durable orchestration
Rt (Red-team) — Not yet implemented
Sm (Small) — Could use Granite / Phi models
Ma (Multi-agent) Agent delegation Elena → Sage → Elena flows
Sy (Synthetic) Story generation Claude/Gemini content creation
Gk (Graph Knowledge) Zep Knowledge Graph Temporal entities + facts
In (Interpret) — Explainability not yet implemented
Th (Thinking) Extended thinking Chain-of-thought in workflows

Gk in Engram: The Implementation

Engram’s Zep integration is a concrete implementation of Graph Knowledge:

graph TD
    A[User Message] --> B[Elena Brain]
    B --> C{Tri-Search}
    C --> D[Semantic Memory]
    C --> E[Knowledge Graph]
    C --> F[Episodic Memory]
    E --> G[Entities]
    E --> H[Facts/Relationships]
    G & H --> I[Context Assembly]
    I --> J[LLM Response]

Key Capabilities

  1. Entity Extraction: Automatic extraction of people, concepts, systems from conversations
  2. Fact Linking: Relationships between entities stored as graph edges
  3. Temporal Awareness: Facts have timestamps, enabling “as of” queries
  4. Cross-Session Learning: Knowledge persists and compounds across sessions

Formal Proposal

We formally propose adding Gk (Graph Knowledge) to the AI Periodic Table:

Symbol: Gk
Name: Graph Knowledge
Position: Row 4 (Emerging), Column 3 (Orchestration)
Definition: Temporal knowledge graphs for dynamic context orchestration
Examples: Zep, Neo4j + LLM, Microsoft GraphRAG

This element completes the Emerging × Orchestration cell and represents the next evolution beyond static frameworks—where orchestration is driven by understanding rather than configuration.

References