Advanced Information

Current Configuration
Setting Value
RAG Mode Normal RAG
Embedding Model Gemini Embedding API
Vector Database Qdrant Cloud
Document Collection simple_rag_docs
Graph Collection simple_rag_graph
Preferred LLM claude
Chunk Size 1000
Chunk Overlap 200
Results Count (Top K) 5
API Rate Limit 60 calls per minute
Embedding Cache Enabled
RAG Mode Comparison
📚 Normal RAG
  • Speed: Fast processing
  • Accuracy: Good for direct facts
  • Use Case: Simple Q&A, document search
  • Storage: Single vector collection
  • Processing: Chunking + embedding
🕸️ Graph RAG
  • Speed: Slower, more thorough
  • Accuracy: Excellent for relationships
  • Use Case: Complex reasoning, connections
  • Storage: Two collections (docs + graph)
  • Processing: Entity extraction + graph building
Performance Features
1. Rate Limiting

SimpleRAG implements rate limiting for API calls to prevent exceeding API provider quotas and avoid service interruptions. This is especially important for Graph RAG which makes additional API calls for entity extraction.

2. Embedding Cache

To improve performance and reduce API costs, SimpleRAG caches embeddings locally. This is particularly beneficial for Graph RAG where similar entities might be processed multiple times.

3. Progress Tracking

For long-running operations like Graph RAG indexing, SimpleRAG provides detailed real-time progress indicators showing entity extraction, relationship mapping, and graph building status.

4. Dual Collection Storage

Graph RAG uses two Qdrant collections: one for document chunks and another for graph elements (entities and relationships), enabling hybrid search strategies.

How It Works
Normal RAG Process
  1. Document parsing and text extraction
  2. Split into overlapping chunks
  3. Generate embeddings using Gemini API
  4. Store in Qdrant vector database
  5. Query with semantic similarity search
  6. Generate answer with Claude LLM
Graph RAG Process
  1. Document parsing and text extraction
  2. Split into larger, context-rich chunks
  3. Extract entities and relationships using Gemini
  4. Build knowledge graph with NetworkX
  5. Generate embeddings for graph elements
  6. Store both docs and graph in Qdrant
  7. Query both collections for hybrid results
  8. Generate context-aware answer with Claude
Graph RAG Technical Details
Entity Extraction

Uses Gemini Pro to identify and categorize entities (PERSON, ORGANIZATION, CONCEPT, LOCATION, EVENT) from document text with descriptions and relationships.

Knowledge Graph Construction

Builds a NetworkX graph where entities are nodes and relationships are edges, enabling graph traversal and neighborhood analysis for enhanced context retrieval.

Hybrid Search Strategy

Combines traditional semantic search of document chunks with graph-based entity and relationship retrieval, providing both direct facts and contextual connections.

Enhanced Prompting

Graph RAG generates specialized prompts that include both document context and relevant entities/relationships, enabling more sophisticated reasoning and answer generation.

About Enhanced SimpleRAG

Enhanced SimpleRAG extends the original system with Graph RAG capabilities, providing two complementary approaches to document Q&A:

  • Normal RAG: Fast, efficient semantic search perfect for direct factual queries
  • Graph RAG: Advanced knowledge graph reasoning ideal for understanding relationships and complex connections

Both modes use the same Gemini API key for embeddings and Claude for answer generation, with Qdrant storing the vector representations. The system automatically handles the complexity of entity extraction, graph construction, and hybrid retrieval strategies.

This dual-mode approach ensures you get the best of both worlds: speed when you need it, and depth when your queries demand sophisticated reasoning about relationships and connections in your documents.