Claude 3.5 Sonnet GraphRAG & Neo4j Knowledge Graph Pipeline

You are a Principal AI Infrastructure Architect specializing in Graph Retrieval-Augmented Generation (GraphRAG), Neo4j knowledge graphs, and hybrid semantic search.

Architect an enterprise GraphRAG ingestion and querying pipeline under the following parameters:
- Domain Entity Space: {Biomedical Clinical Trials, Drug Compounds, Gene Targets, and Adverse Reactions}
- Knowledge Graph Store: {Neo4j AuraDB Enterprise with Vector Indexing}
- Extraction Chunking Strategy: {Hierarchical Leiden Community Detection (Microsoft GraphRAG approach)}
- Retrieval Mode: {Two-Stage: Dense Vector Cosine Seed + 2-Hop Relationship Expansion}

### ARCHITECTURAL REQUIREMENTS:
1. **Graph Ontology & Schema Design**:
   - Define strict Node Labels, Edge Types (Relationships), and Properties.
   - Ensure directional semantic consistency (e.g., `(:Company)-[:ACQUIRED {date, valuation}]->(:Company)`).
2. **Entity & Relation Extraction Prompt**:
   - Provide a zero-shot Claude 3.5 Sonnet extraction prompt that extracts high-order relations and outputs deterministic JSON.
3. **Hybrid Cypher & Vector Retrieval Pipeline**:
   - Implement Python code combining Neo4j Vector Index (`db.index.vector.queryNodes`) with multi-hop Cypher path traversal (`CALL apoc.path.subgraphNodes`).
4. **Context Window Synthesis Engine**:
   - Format both the structured graph sub-topology and text chunk embeddings into a coherent prompt payload for the generation LLM.