autograph-llmneo4j 0.1.0

Dataset-agnostic knowledge graph builder with optional LLM and Neo4j integrations.

Auto Graph Builder Framework

Transform any dataset into a Knowledge Graph with just a few lines of code

---

🎯 Vision

A framework that automatically converts structured data (JSON, CSV, Excel) into optimized Neo4j graph databases using LLM-powered entity extraction and optional embeddings.

Goal: Enable developers to build knowledge graphs without graph expertise.

---

💡 The Problem

Building a graph database today requires:

# Current approach: 100+ lines of code + expertise

1. Manually design graph schema
   - Which nodes? Which relationships?
   - What properties to store?
   
2. Write entity extraction logic
   - Parse text fields
   - Extract organizations, people, locations
   - Handle edge cases
   
3. Create database schema
   - Write Cypher for constraints
   - Create indexes for performance
   - Test and debug
   
4. Insert data
   - Write complex Cypher queries
   - Handle batching
   - Error handling
   
5. Add embeddings (optional)
   - Generate vectors
   - Create vector indexes
   - Sync with graph

6. Optimize & maintain
   - Monitor performance
   - Add new indexes
   - Update schema

Result: Takes days/weeks, requires Neo4j expertise, error-prone.

---

✨ Our Solution

# Auto Graph approach: 3 lines

from autograph import GraphBuilder

builder = GraphBuilder(
    llm_api_key="sk-...",           # OpenAI or Anthropic
    neo4j_uri="bolt://localhost:7687",
    neo4j_password="password",
    use_embeddings=True              # Optional
)

builder.load_data("data.json").build_graph()

Result: Working knowledge graph in minutes, zero graph expertise needed.

What Happens Automatically

1. 🤖 LLM analyzes your data structure
   → Designs optimal graph schema
   → Suggests node types & relationships

2. 🧠 LLM extracts entities from text
   → Organizations, people, locations, etc.
   → Creates connections automatically

3. ⚡ Framework optimizes database
   → Creates constraints for uniqueness
   → Adds indexes for performance
   → Batch processing for efficiency

4. 📊 Optional: Generates embeddings
   → Vector representations of text
   → Enables semantic search
   → Hybrid graph + vector queries

---

🌟 Core Capabilities

1. Zero Configuration

• No manual schema design
• No Cypher knowledge required
• Works out of the box

2. LLM-Powered Intelligence

• Automatic schema detection
• Smart entity extraction
• Relationship inference

3. Multi-Format Support

• JSON files
• CSV spreadsheets
• Excel workbooks
• Pandas DataFrames

4. Optional Semantic Search

• Toggle embeddings on/off
• Automatic vector indexing
• Hybrid graph + semantic queries

5. Production Ready

• Auto-optimization
• Batch processing
• Error handling
• Performance monitoring

---

🏗️ How It Works

┌──────────────────────────────────────────────────────┐
│              YOUR DATA (JSON/CSV/Excel)              │
└────────────────────┬─────────────────────────────────┘
                     │
                     ▼
┌──────────────────────────────────────────────────────┐
│                 AUTO GRAPH BUILDER                   │
│                                                      │
│  Step 1: Analyze Data                               │
│  ├─ LLM examines structure                          │
│  └─ Designs graph schema                            │
│                                                      │
│  Step 2: Extract Entities                           │
│  ├─ LLM reads text fields                           │
│  └─ Identifies entities & relationships             │
│                                                      │
│  Step 3: Build Graph                                │
│  ├─ Create nodes & relationships                    │
│  ├─ Add constraints & indexes                       │
│  └─ Optimize for queries                            │
│                                                      │
│  Step 4: Optional Embeddings                        │
│  ├─ Generate vectors                                │
│  └─ Create vector index                             │
└────────────────────┬─────────────────────────────────┘
                     │
                     ▼
┌──────────────────────────────────────────────────────┐
│           NEO4J KNOWLEDGE GRAPH (Ready!)             │
└──────────────────────────────────────────────────────┘

Key Components

1. Data Loaders - Understand your data format
2. Schema Analyzer - LLM designs graph structure
3. Entity Extractor - LLM finds entities in text
4. Graph Manager - Builds optimized Neo4j database
5. Embedding Generator - Optional semantic search

---

🚀 Quick Start

Basic Usage

from autograph import GraphBuilder

# Initialize
builder = GraphBuilder(
    llm_api_key="sk-...",
    neo4j_uri="bolt://localhost:7687",
    neo4j_password="password"
)

# Build graph automatically
builder.load_data("data.json").build_graph()

# Search
results = builder.search("AI regulation")

With Semantic Search

builder = GraphBuilder(
    llm_api_key="sk-...",
    neo4j_uri="bolt://localhost:7687",
    neo4j_password="password",
    use_embeddings=True  # Enable vector search
)

builder.load_data("data.json").build_graph()

# Semantic search automatically enabled
results = builder.search("government AI policy")

Different Data Formats

# JSON
builder.load_data("news.json")

# CSV
builder.load_data("products.csv")

# Excel
builder.load_data("customers.xlsx")

# All trigger automatic processing
builder.build_graph()

---

🎨 Example Use Cases

News & Media

Transform RSS feeds and articles into searchable knowledge graphs with automatic entity extraction and topic clustering.

E-Commerce

Build product catalogs with automatic category hierarchies, brand relationships, and customer review connections.

Research & Academia

Create citation networks from paper databases with author affiliations and topic relationships.

Social Networks

Map user interactions, followers, and content sharing patterns automatically from platform exports.

Business Intelligence

Convert CRM data into relationship graphs showing customer journeys, sales patterns, and market segments.

---

⚙️ Configuration

Minimal Setup

builder = GraphBuilder(
    llm_api_key="sk-...",
    neo4j_uri="bolt://localhost:7687",
    neo4j_password="password"
)

Common Options

builder = GraphBuilder(
    # Required
    llm_api_key="sk-...",
    neo4j_uri="bolt://localhost:7687",
    neo4j_password="password",
    
    # Optional
    use_embeddings=True,        # Enable semantic search
    llm_provider="openai",      # or "anthropic"
    verbose=True                # Show progress
)

---

🤖 How LLM Powers the Framework

Automatic Schema Detection

The framework sends sample data to LLM:

Input: Sample of your data

[
  {
    "title": "White House considers AI regulation",
    "source": "NYT",
    "content": "The White House is discussing..."
  }
]

LLM Analyzes and Returns:

{
  "node_types": [
    {"label": "Article", "id_field": "url"},
    {"label": "Source", "id_field": "name"},
    {"label": "Entity", "id_field": "name"}
  ],
  "relationships": [
    {"type": "PUBLISHED_BY", "from": "Article", "to": "Source"},
    {"type": "MENTIONS", "from": "Article", "to": "Entity"}
  ]
}

Framework creates the graph automatically based on this schema.

---

Automatic Entity Extraction

For each record, LLM extracts entities:

Input Text:

"White House considers vetting AI models. 
The Biden administration is exploring..."

LLM Extracts:

{
  "organizations": ["White House", "Biden administration"],
  "technologies": ["AI models"],
  "people": []
}

Framework creates nodes and relationships for these entities.

---

⚡ Performance & Costs

Processing Time Estimates

Dataset Size	Processing Time	With Embeddings
100 records	~2 minutes	~5 minutes
1,000 records	~15 minutes	~40 minutes
10,000 records	~2 hours	~6 hours

Using GPT-4 and standard embedding models

Cost Estimates (OpenAI)

For 1,000 records:

• Schema analysis: ~$0.01
• Entity extraction: ~$2.00
• Embeddings (optional): ~$0.02
• Total: ~$2.03

Neo4j Requirements

• Free tier: 200MB (suitable for ~10-50k articles)
• Production: Aura Pro $65/month (8GB+)

---

🎯 Target Users

Primary:

• Data Scientists who know Python but not Neo4j
• Backend Developers building features quickly
• ML Engineers needing knowledge graphs for RAG

Secondary:

• Researchers creating quick prototypes
• Startups building MVPs fast
• Students learning about graphs

---

💎 Value Proposition

For Developers

• ✅ 90% less code required
• ✅ Zero graph expertise needed
• ✅ Production-ready in minutes

For Businesses

• ✅ Faster time to market
• ✅ Lower development costs
• ✅ Easier maintenance

For Projects

• ✅ Rapid prototyping
• ✅ Easy experimentation
• ✅ Scalable foundation

---

🚀 Getting Started

# Install (when available)
pip install auto-graph

# Use
from autograph import GraphBuilder

builder = GraphBuilder(
    llm_api_key="sk-...",
    neo4j_uri="bolt://localhost:7687",
    neo4j_password="password"
)

builder.load_data("your_data.json").build_graph()

---

📝 Summary

Auto Graph Builder transforms any dataset into a knowledge graph with minimal code.

• Zero configuration - LLM handles schema design
• Smart extraction - Entities detected automatically
• Production ready - Optimized and scalable
• Developer friendly - Works like familiar Python libraries

Built for developers who want graphs, not graph expertise.