Ultimate RAG (Hybrid CRAG + Self-RAG)

Advanced retrieval-augmented generation combining Corrective RAG and Self-RAG techniques with mathematical convergence guarantees.

Architecture

Query → [Vector Search] → [Relevance Check] → [Self-Critique] → [Corrective Loop] → Answer
                              ↓                    ↓
                         [Web Search]        [Regenerate]
                              ↓                    ↓
                         [Knowledge Graph] ← [Lyapunov Check]

Features

Corrective RAG (CRAG)

• Relevance scoring with threshold-based correction
• Automatic web search fallback for low-confidence retrievals
• Knowledge graph augmentation for entity relationships

Self-RAG

• Self-critique mechanism for answer validation
• Iterative refinement with convergence tracking
• Hallucination detection and prevention

Lyapunov Convergence

• Mathematical proof of answer stability
• Guaranteed convergence within N iterations
• Quantifiable improvement metrics

Usage

from AFO.rag import UltimateRAG

rag = UltimateRAG(top_k=5, max_iterations=10)
result = rag.query("What is the Trinity Score philosophy?")

# Result includes confidence and convergence metrics
print(f"Answer: {result.answer}")
print(f"Confidence: {result.confidence}")
print(f"Iterations: {result.iterations}")
print(f"Lyapunov Delta: {result.lyapunov_delta}")

Parameters

Philosophy Alignment

• 眞 (Truth): Lyapunov-proven retrieval accuracy
• 善 (Goodness): No hallucinations, stable answers
• 美 (Beauty): Clean retrieval pipeline
• 孝 (Serenity): Auto-converges without user intervention