vrin 0.2.3

Optimized Hybrid RAG SDK with smart deduplication, enhanced fact extraction, and competitive performance

VRIN Hybrid RAG SDK v0.2.1

Optimized Hybrid RAG SDK with smart deduplication, enhanced fact extraction, and competitive performance.

🚀 New in v0.2.1 - Performance Optimizations

• 🎯 Smart Deduplication - Prevents duplicate facts and chunks, reduces storage by 40-60%
• ⚡ Enhanced Performance - Sub-3s queries with 7+ combined graph + vector results
• 🧠 Improved Fact Quality - 0.8+ confidence facts with self-updating system
• 💾 Storage Optimization - Only stores unique, fact-rich content
• 🔄 Self-Improving - Higher confidence facts automatically update existing ones
• 📊 Competitive Edge - Outperforms pure RAG and basic GraphRAG systems

🚀 Core Features

• ⚡ Hybrid RAG Architecture - Graph reasoning + Vector similarity search
• 🧠 Intelligent Entity Matching - AI-powered compound entity recognition
• 📊 Advanced Fact Extraction - High-confidence structured knowledge extraction
• 🔍 Sub-3s Query Response - Optimized retrieval with comprehensive coverage
• 🎯 AI-Powered Summaries - Natural language answers with cited sources
• 📈 Enterprise-Ready - User isolation, authentication, and production scaling

📦 Installation

pip install vrin

🔧 Quick Start

from vrin import VRINClient

# Initialize with your API key
client = VRINClient(api_key="your_vrin_api_key")

# Insert knowledge with automatic fact extraction
result = client.insert(
    content="Python is a high-level programming language created by Guido van Rossum in 1991. It emphasizes code readability and supports multiple programming paradigms.",
    title="Python Programming Language",
    tags=["programming", "python", "language"]
)

print(f"✅ Extracted {result['facts_extracted']} facts")
# Output: ✅ Extracted 7 facts

# Query with intelligent hybrid search
response = client.query("Who created Python and when?")
print(f"📝 Answer: {response['summary']}")
print(f"⚡ Performance: {response['total_facts']} graph + {response['total_chunks']} vector = {response['combined_results']} results")
print(f"🔍 Query time: {response['search_time']}")

# Output: 
# 📝 Answer: Python was created by Guido van Rossum in 1991. It is a high-level programming language known for its simplicity and readability...
# ⚡ Performance: 4 graph + 5 vector = 9 results
# 🔍 Query time: 2.85s

🔍 Advanced Usage

Storage Optimization Features

# Insert content with optimization details
result = client.insert(content, title, tags)

print(f"🧠 Facts extracted: {result['facts_extracted']}")
print(f"💾 Storage optimization: {result['storage_optimization']}")
print(f"📦 Chunk stored: {result['chunk_stored']}")
print(f"⚡ Processing time: {result['processing_time']}")

# Check deduplication efficiency
storage = result['storage_result']
print(f"📈 New facts: {storage['stored_facts']}")
print(f"🔄 Updated facts: {storage['updated_facts']}")
print(f"⏭️ Skipped duplicates: {storage['skipped_duplicates']}")
print(f"💾 Efficiency: {storage['storage_efficiency']}")

Raw Results Access

# Get detailed graph facts and vector results
details = client.get_raw_results("Python programming language")

print("📊 Graph Facts:")
for fact in details['graph_facts']:
    confidence = fact['confidence']
    print(f"  • {fact['subject']} → {fact['predicate']} → {fact['object']} (confidence: {confidence:.2f})")

print(f"\n🔎 Vector Chunks: {len(details['vector_results'])}")
print(f"🧠 Entities Found: {details['entities_found']}")

Knowledge Graph Visualization

# Get knowledge graph data for visualization
graph = client.get_knowledge_graph()
print(f"📈 Graph: {len(graph['data']['nodes'])} nodes, {len(graph['data']['edges'])} edges")

Batch Processing

# Insert multiple documents efficiently
documents = [
    {"content": "Machine learning is a subset of AI...", "title": "ML Basics"},
    {"content": "Neural networks consist of layers...", "title": "Neural Networks"},
    {"content": "Deep learning uses multiple layers...", "title": "Deep Learning"}
]

for doc in documents:
    result = client.insert(doc['content'], doc['title'])
    print(f"Processed: {doc['title']} -> {result['facts_extracted']} facts")

🎯 API Reference

VRINClient

__init__(api_key: str)

Initialize the VRIN client with your API key.

insert(content: str, title: str = None, tags: List[str] = None) -> Dict

Insert knowledge into the system with automatic fact extraction.

Returns:

• success: Whether the operation succeeded
• facts_extracted: Number of facts extracted
• chunk_id: Unique identifier for the content
• message: Status message

query(query: str, include_summary: bool = True, include_raw_results: bool = False) -> Dict

Query the knowledge base with optimized hybrid retrieval.

Parameters:

• query: Search query string
• include_summary: Include AI-generated summary (default: True)
• include_raw_results: Include detailed graph facts and vector results

Returns:

• success: Whether the query succeeded
• summary: AI-generated comprehensive answer
• search_time: Query execution time
• entities_found: Extracted entities from the query
• total_facts: Number of graph facts found
• total_chunks: Number of vector chunks found

get_raw_results(query: str) -> Dict

Get complete raw results including graph facts and vector chunks.

get_knowledge_graph() -> Dict

Get knowledge graph visualization data with nodes and edges.

🔐 Authentication

1. Sign up at VRIN Console
2. Create a new API key
3. Use the API key to initialize your client

client = VRINClient(api_key="vrin_your_api_key_here")

🏗️ Architecture

VRIN uses a sophisticated Hybrid RAG architecture:

1. Fact Extraction - LLM-powered extraction of structured facts
2. Graph Storage - Facts stored as a knowledge graph in Neptune
3. Vector Storage - Semantic embeddings in OpenSearch
4. Hybrid Retrieval - Combines graph traversal + vector similarity
5. Result Fusion - Intelligent ranking and result combination
6. AI Summarization - Natural language response generation

📊 Performance (v0.2.1 Optimized)

• Fact Extraction: 3-8 high-quality facts per insertion (0.8+ confidence)
• Query Response: Sub-3s with 7+ combined graph + vector results
• Hybrid Coverage: 2-5 graph facts + 3-5 vector chunks per query
• Storage Efficiency: 40-60% reduction through smart deduplication
• Self-Improvement: Facts automatically update with higher confidence versions
• Competitive Advantage: Outperforms single-method RAG systems

🛠️ Requirements

• Python 3.8+
• Active internet connection
• VRIN API key

📄 License

MIT License - see LICENSE file for details.

🤝 Support

• 📧 Email: [e-mail]
• 📚 Documentation: https://docs.vrin.ai
• 🐛 Issues: https://github.com/vrin-ai/vrin-sdk/issues

---

Built with ❤️ by the VRIN Team