Skip to content

Ragatanga Architecture

This document provides a technical overview of Ragatanga's architecture, components, and design decisions.

System Overview

Ragatanga is a hybrid retrieval system that combines ontology-based reasoning with semantic search. This combination provides several advantages:

  1. Structure + Unstructured: Combines the precision of ontology/SPARQL queries with the flexibility of semantic search
  2. Adaptive Parameters: Adjusts retrieval strategies based on query characteristics
  3. Extensible Design: Supports multiple embedding and LLM providers
  4. Confidence Scoring: Ranks results by confidence for better answers

Architecture Diagram

┌─────────────────────────────────────────────────────────────┐
│                      Ragatanga System                       │
└──────────────────────────────┬──────────────────────────────┘
                               │
          ┌───────────────────┼───────────────────┐
          │                   │                   │
┌─────────▼─────────┐ ┌───────▼──────┐   ┌────────▼────────┐
│      Core         │ │    Config    │   │     Version     │
└─────────┬─────────┘ └──────────────┘   └─────────────────┘
          │                                        
┌─────────▼─────────┐                   ┌────────────────────┐
│  ┌─────────────┐  │                   │        API         │
│  │  Ontology   │◄─┼───────────────────┤                    │
│  └─────┬───────┘  │                   │  ┌─────────────┐   │

Core Components

Ontology Manager

The OntologyManager is responsible for:

  • Loading and parsing ontology files (TTL format)
  • Materializing inferences using RDFS/OWL reasoning
  • Providing SPARQL query capabilities
  • Managing the knowledge graph

Adaptive Retriever

The AdaptiveRetriever is the central component that:

  • Analyzes queries to determine the best retrieval strategy
  • Combines ontology-based and semantic search results
  • Ranks and scores results based on confidence
  • Provides a unified interface for all retrieval operations

Embedding Manager

The EmbeddingManager handles:

  • Text embedding generation using various providers
  • Vector storage and retrieval
  • Similarity search operations
  • Caching of embeddings for performance

LLM Manager

The LLMManager provides:

  • Integration with multiple LLM providers
  • Query enhancement and reformulation
  • Structured answer generation
  • Confidence estimation

Data Flow

  1. Query Analysis: Incoming queries are analyzed to determine complexity and type
  2. Retrieval Strategy Selection: Based on analysis, the system selects appropriate retrieval methods
  3. Parallel Execution: Ontology and semantic search operations run in parallel
  4. Result Fusion: Results from different sources are combined and ranked
  5. Answer Generation: Final answers are generated with confidence scores

Extension Points

Ragatanga is designed to be extensible in several ways:

  • Custom Embedding Providers: Add support for new embedding models
  • Custom LLM Providers: Integrate with additional LLM services
  • Custom Retrieval Strategies: Implement specialized retrieval methods
  • Custom Scoring Functions: Define domain-specific relevance scoring