Advanced Intelligence Pipelines
Move beyond simple retrieval to autonomous strategic reasoning and complex investigative workflows.
The Pometry ecosystem enables the construction of high-order intelligence pipelines that don't just "find data" - they reason about it. These pipelines leverage the temporal graph as a living world model, allowing agentic personas to perform multi-step analysis, self-correct their queries, and generate deep narrative insights.
Commercial Capability: The following examples showcase advanced features built using Pometry's managed Agentic APIs and enterprise UI components. These pipelines represent the target capability achievable by combining Raphtory's temporal engine with Pometry's intelligence orchestration.
Intelligence Personas
Pometry pipelines are often organized into specialized "Intelligence Personas," each utilizing different dimensions of the temporal graph.
The Historian
Specializes in temporal causality. Traces the evolution of an entity over years to detect "Long-Game" social engineering or gradual risk accumulation.
The Forecaster
Predictive intelligence. Uses temporal graph metrics and FastRP embeddings to identify nodes showing symptoms of future failure or systemic bottlenecking.
The Investigator
Pattern-based matching. Deploys complex multi-hop logic to flag schemes like circular transactions, scatter-gather, and account layering automatically.
The Investigator: Pattern Matching
Investigative pipelines use Raphtory's temporal DFS (Depth First Search) to identify specific structural signatures. Unlike SQL or standard Graph DBs, Pometry can match patterns that are separated by both hops and time.
Detecting Circular Transactions
This pipeline identifies "Round-Tripping" - where capital leaves an entity and returns via a convoluted chain of intermediaries within a specific temporal window.
The Classifier: Entity Resolution (ER)
One of the most powerful pipelines is the Temporal Identifier. It merges disparate data sources into a unified entity by analyzing structural similarity over time.
Classification Logic
By comparing the temporal "fingerprint" of two nodes (who they talk to, when, and how frequently), Pometry can classify them as the same real-world entity even if their metadata (Name, Email) slightly differs.
Personas in Action
The Historian: Temporal Lineage
The Historian uses Raphtory's event-sourcing to re-verify the "Source of Wealth" by looking at the entire history of an entity's connections.
The Investigator: Advanced Schemes
The Investigator deploys multi-hop DFS protocols to catch coordinated financial crimes.
1. Scatter-Gather (The Funnel)
Detects when a single source "scatters" small amounts to multiple intermediaries who then "gather" the consolidated sum to a final destination.
2. Dormant Activation
Flags accounts that have been inactive for long periods (e.g., >120 days) and suddenly execute high-value outbound transfers.
The Forecaster: Predictive Symptoms
The Forecaster analyzes structural changes - like a sudden spike in Temporal Degree Centrality or a shift in FastRP Embedding space - to predict an entity's likelihood of being involved in a future risk event.
Semantic Triplet & Similarity APIs
Pometry provides specialized APIs for grounding AI in the relationships (edges) rather than just the objects (nodes).
The Edge Triplet context
When an AI agent searches for "suspicious interactions," the Pometry Vector API returns Edge Triplets: (Source) --[Semantic Context]--> (Destination). This ensures the AI understands the nature of the link, not just its existence.
Investigator, find any "unusual financial handshakes" between the shell company and the sanctioned entities.
Agentic UI: Rendering the Thought Process
Pometry's AiChat components allow the AI to show its "work" as it navigates the graph. This creates a transparent audit trail for compliance.
- Thought Traces: The AI shows which Python code it is about to run.
- Execution Logs: Real-time feedback from the Raphtory engine.
- Grounded Visuals: Automatically generated subgraphs that prove the narrative.
Next Steps
High-Order Matching & ER
Deep dive into structural similarity and entity resolution.
Agentic RAG (Python)
Build tools that write graph code for your agents.
Core Capabilities Comparison
| Feature | Standard RAG | Pometry Advanced Pipelines |
|---|---|---|
| Agency | Passive Retrieval | Autonomous Tool Use (Python/GQL) |
| Logic | Fuzzy Similarity | Hard Temporal & Structural Logic |
| Output | Text Summary | Actionable Intelligence & Decision Support |
| Traceability | Black Box | Visual Graph Grounding (Minimaps) |