WtrDB for Insurance
Turning policy documents into actionable intelligence — a federated, queryable knowledge graph backed by mathematical quality guarantees.
Insurance knowledge is locked in documents.
Insurance companies sit on massive volumes of unstructured documents — policy wordings, claims files, underwriting guidelines, regulatory filings, reinsurance treaties, loss run reports, and actuarial memos. These documents contain critical knowledge about risk relationships, coverage dependencies, and regulatory obligations, but that knowledge is locked in silos. Traditional approaches (keyword search, basic RAG) return the nearest paragraph. They can't answer questions that require traversals across multiple documents — like tracing how a specific exclusion clause in a reinsurance treaty affects a downstream policyholder's claim.
From paragraphs to relationships.
WtrDB transforms unstructured insurance documents into a federated, queryable knowledge graph backed by mathematical quality guarantees. Instead of returning paragraphs, it lets AI agents walk relationships between entities — policies, clauses, claimants, coverages, risks, regulations — and assemble answers from facts distributed across hundreds of documents.
Four steps from documents to intelligence
Ingest
Drop in policy documents, claims files, regulatory filings (.pdf, .md, .txt, .csv). No pre-processing required.
Extract
A three-stage LLM pipeline pulls entities (insurers, policyholders, coverages, perils, exclusions, limits) and relationships, then verifies every extracted fact against the source text. Nothing hallucinates into the graph.
Federate
Multiple knowledge bases (underwriting, claims, compliance) are merged using mathematically governed conflict resolution. Cross-department contradictions are detected and surfaced, not silently overwritten.
Query
Expose the graph as an MCP endpoint that any AI agent (Claude, Cursor, custom tools) can query instantly. No SDK. No integration work.
Explore and rotate your WtrDB graph in 3D
Drag to rotate. Scroll to zoom. Right-click and drag to pan. Click a node to inspect its relationships.
Six applications across the insurance value chain.
Claims Processing & Triage
- Cross-reference claims against policy terms automatically. An agent can walk from a claim event to the applicable policy, check coverage limits, identify relevant exclusions, and flag subrogation opportunities — across documents that were never explicitly linked.
- Detect duplicate or related claims by traversing entity relationships rather than relying on string matching.
- Reduce claims leakage by surfacing coverage terms and exclusion clauses that adjusters might miss in a 200-page policy.
Underwriting Intelligence
- Build a living map of risk relationships. WtrDB's 13 entity types and 25 relationship types capture the full topology of insured risks, coverages, exclusions, endorsements, and their dependencies.
- Query across the entire book of business. "Which policies have flood exposure in these ZIP codes with limits above $5M?" becomes a graph traversal, not a manual file review.
- Track how endorsements and amendments ripple through coverage structures — the graph preserves temporal evolution, so you can see what changed and when.
Regulatory Compliance & Audit
- Map regulatory requirements to internal controls and evidence. WtrDB includes built-in compliance frameworks (SOC 2, HIPAA, CMMC) and automated evidence collection. For insurance, this extends to state filing requirements, NAIC guidelines, and Solvency II obligations.
- Produce audit-ready evidence trails. Every fact in the graph links back to its source document and chunk. Governance adjudication ensures nothing enters the graph without evidence verification.
- Detect compliance gaps. The sheaf-theoretic consistency layer identifies contradictions — for example, where internal policy guidelines conflict with filed regulatory requirements.
Reinsurance & Treaty Management
- Federate insurer and reinsurer knowledge bases. WtrDB's federation layer aligns entities across separate graphs and resolves conflicts mathematically — critical when reconciling cedant and reinsurer views of the same risk.
- Trace coverage chains. Walk from a loss event through the primary policy, to the applicable reinsurance treaty, through to retrocession — across documents owned by different parties.
- Monitor treaty consistency. The sheaf Laplacian spectral analysis continuously measures whether the federated view of shared risk is internally consistent, flagging discrepancies before they become disputes.
Fraud Detection & Investigation
- Surface hidden entity relationships that text search would never find. The graph reveals connections between claimants, providers, adjusters, and loss events that span separate files and time periods.
- Temporal evolution tracking. WtrDB's dual-track extraction distinguishes static facts from evolutionary events, preserving the full history of how entities and relationships changed over time — critical for identifying patterns in serial fraud.
- Conservative contradiction handling. Nothing is deleted — deprecated facts are soft-marked, preserving the complete evidential record for investigations.
Agent-Powered Customer Service
- Give customer-facing AI agents a real memory. Wire a WtrDB Brain into your agent, and it can answer policyholder questions by reasoning over actual policy terms, coverage structures, and claims history — not just regurgitating the nearest FAQ paragraph.
- MCP-native. Works out of the box with Claude Desktop, Cursor, and any MCP-compatible agent framework. Per-endpoint API keys with rotate/revoke controls for security.
Built for regulated, document-heavy industries.
Mathematical Quality Guarantees
Unlike typical RAG systems, WtrDB uses cellular sheaf theory to continuously measure and maintain consistency across knowledge graphs. Sheaf Laplacian analysis provides a single consistency score across the entire graph — a quantitative answer to "how reliable is our knowledge base?" Cohomology-based conflict detection identifies structural contradictions (e.g., a policy that both covers and excludes the same peril under different conditions) as independent conflict cycles. Sheaf diffusion automatically smooths inconsistencies while preserving the underlying structure. For a regulated industry where decisions must be defensible, this is not academic — it's operational.
Governed Knowledge Evolution
Insurance data changes constantly: endorsements amend policies, claims progress through stages, regulations update. WtrDB handles this through incremental extraction (only processes new or changed documents), three-filter governance (evidence verification, logical verification, and evolutionary-intent verification ensure updates are valid before entering the graph), and soft deprecation over hard deletion — historical states are preserved for audit, not overwritten.
Enterprise-Grade Access Control
RBAC with team scoping — underwriters see underwriting graphs, claims teams see claims graphs, compliance sees everything. Full audit trail — every graph modification, every query, every access event is logged. MFA and OPA policy engine — meets CMMC and SOC 2 requirements out of the box.
Data Stays on Your Infrastructure
WtrDB is deployed within your own environment. No policy documents, PII, PHI, or proprietary underwriting models leave your network — a non-negotiable for regulated insurers.
Insurance runs on documents. WtrDB makes them queryable.
Faster claims, smarter underwriting, airtight compliance, and AI agents that actually understand your business — not just the nearest paragraph.