Elsai Graph

Elsai Graph is a toolkit for building, storing, and querying knowledge graphs powered by LLMs. It bridges unstructured text and structured knowledge — enabling graph-powered RAG pipelines that understand relationships, not just similarity.

Architecture

Packages

Package	Version	Purpose
elsai-graph-constructor	0.1.0	Extract entities and relationships from text
elsai-graph-generator	0.1.0	Store graphs in Neo4j; manage vector indexes
elsai-graph-query-rag	0.1.0	Answer natural language questions from the graph

Requirements: Python >= 3.10, Neo4j instance (for generator and query-rag)

Quick end-to-end example

# Step 1 — Extract graph from text
from elsai_graph_constructor import GraphConstructor

constructor = GraphConstructor(llm=llm)
success, message, graph_data = constructor.construct_graph(chunks=[
    {"chunk_id": "c1", "text": "Alice works at Acme Corp in Paris."},
    {"chunk_id": "c2", "text": "Acme Corp is a leader in aerospace technology."},
])

# Step 2 — Store in Neo4j
from elsai_graph_generator import Neo4jConnector, GraphStorage, EmbeddingManager

connector = Neo4jConnector(uri="bolt://localhost:7687", user="neo4j", password="password")
connector.connect()

storage = GraphStorage(connector)
storage.store_graph(graph_data)

embedding_manager = EmbeddingManager(connector, embedding_model)
embedding_manager.create_vector_index("entity_embeddings")
embedding_manager.embed_nodes()

# Step 3 — Query the graph
from elsai_graph_query_rag import Neo4jRetriever, HybridGraphRetriever, GraphRAGEngine

neo4j_retriever = Neo4jRetriever(connector, llm)
hybrid_retriever = HybridGraphRetriever(connector, embedding_manager, neo4j_retriever)
rag_engine = GraphRAGEngine(neo4j_retriever, hybrid_retriever, llm)

success, message, answer = rag_engine.ask("Where does Alice work?")
print(answer)

Next steps

• Graph Construction — entity & relationship extraction
• Graph Storage — Neo4j storage and vector indexing
• Graph RAG — answering questions from the graph