Back to Portfolio
Live ProjectEconomic Development

Enterprise Relationship Graph

GraphRAG-powered knowledge graph mapping 500+ business entities and 2,000+ relationships across the Jacksonville ecosystem. LLM-driven entity extraction, Leiden community detection, and three search paradigms (local, global, drift) enable multi-hop discovery of partnership networks, funding flows, and industry clusters.

500+
Entities Mapped
2,000+
Relationships
3 paradigms
Search Modes

Challenge

Jacksonville's business ecosystem has thousands of organizations, but no unified way to understand how they connect — partnerships, shared board members, investment relationships, and industry clusters are invisible. Traditional search returns isolated facts without revealing the multi-hop relationship paths between entities.

Approach

Built a GraphRAG pipeline: 1. Ingested public business data from multiple sources. 2. Used LLM-powered entity extraction (OpenAI GPT-4) to identify organizations, people, programs, and funding relationships. 3. Applied Leiden community detection to discover thematic clusters — Healthcare, Financial Services, Logistics, Innovation Programs, Investment, and Leadership. 4. Deployed three retrieval paradigms: Local Search for entity traversal, Global Search across community summaries, and Drift Search for iterative multi-hop discovery.

Outcome

500+ business entities mapped with 2,000+ relationships across 6 detected communities. Economic development professionals use multi-hop graph traversal to discover partnership opportunities invisible to keyword search — e.g., tracing a 4-hop path from investment funding through accelerator programs to healthcare innovation. The graph auto-updates as new data sources are ingested.

Architecture

Data Ingestion (Public Records, APIs)
Entity Extraction (OpenAI GPT-4)
Community Detection (Leiden)
Graph Database (Neo4j)
Visualization (D3.js Force Graph)
Search (Local / Global / Drift)
Auto-Update Pipeline

Interactive Demo

GraphRAG Pipeline

DocumentsPublic records, APIs
ChunkingSemantic segmentation
Entity ExtractionGPT-4 NER
Knowledge GraphNeo4j storage
Community DetectionLeiden algorithm
RetrievalLocal / Global / Drift

Three Search Paradigms

Local Search

Traverses entity neighborhoods in the graph — e.g., "Who are Mayo Clinic's partners?" follows direct edges to related organizations, programs, and people.

"Which programs does Baptist Health sponsor?"

Global Search

Queries community-level summaries generated by Leiden detection — ideal for broad thematic questions spanning the entire ecosystem.

"What are Jacksonville's key industry clusters?"

Drift Search

Iteratively refines queries across community boundaries, following multi-hop paths to uncover deep, non-obvious connections.

"How does venture funding reach healthcare innovation?"

Technology Stack

Next.jsOpenAIGraphRAGNeo4jD3.jsNode2VecLeiden Algorithm

Book Your AI Consultation

Start with a free consultation. We'll assess your AI readiness, identify high-impact opportunities, and scope a concrete first engagement.

+1
0 / 2000