Architecture¶
Comprehensive documentation for Ryoma’s architecture and design patterns.
Platform Architecture¶
The platform is organized into three distinct packages with clear separation of concerns:
ryoma_data - Data layer with connectors and profiling
ryoma_ai - AI layer with LLM agents and analysis
ryoma_lab - UI layer with interactive interfaces
Enhanced SQL Agent¶
The Enhanced SQL Agent combines cutting-edge research with enterprise reliability for Text-to-SQL tasks.
Key Features¶
Multi-step reasoning with intelligent workflow
Advanced schema linking algorithms
Comprehensive safety validation
Intelligent error handling with auto-recovery
ReFoRCE optimizations for state-of-the-art performance
Database Profiling System¶
Comprehensive metadata extraction based on research from “Automatic Metadata Extraction for Text-to-SQL”.
Key Features¶
Statistical analysis (row counts, NULL stats, distinct-value ratios)
Type-specific profiling (numeric, date, string)
Semantic type inference (emails, phones, URLs)
Data quality scoring
LSH-based column similarity
Top-k frequent values
Store Architecture¶
Unified storage system separating metadata, vectors, and data sources.
Three-Store Architecture¶
Metadata Store - Structured metadata, configuration, and state
Vector Store - Semantic search and embeddings
Data Sources - Actual database connections