Acuario - RAG (Retrieval Augmented Generation) Agent

Overview

Acuario is the agent responsible for implementing RAG (Retrieval Augmented Generation) capabilities in the TKM AI Agency Platform. It performs semantic search across document and image embeddings to provide context-aware responses.

Directory Structure

Backend/CRM/Acuario/
├── acuario.py         # Main agent implementation
├── acuario_tools.py   # RAG utilities and search functions
└── acuario_schema.py  # Data models and schemas

Main Components

AcuarioAgent Class

The main class that handles RAG operations:

Embedding-based search
Context retrieval
Response generation
Integration with other agents

Search Pipeline

Query embedding retrieval
Multi-modal content search
Context aggregation
Response generation

Key Features

Semantic Search

Vector similarity search
Multi-modal content support
Configurable similarity threshold
Performance optimization

Content Types

Document embeddings
Image embeddings
Combined search results
Metadata preservation

Integration Features

Database session management
Performance monitoring
Error handling
Metadata tracking

API Operations

RAG Request

# Request Format
{
    "embedding": List[float],
    "original_query": str
}
 
# Response Format
{
    "success": bool,
    "response": str,
    "context": List[dict],
    "metadata": {
        "processed_at": str,
        "sources_used": {
            "document_sources": int,
            "image_sources": int
        }
    }
}

Integration

Database Integration

SingleStore vector operations
Embedding tables:
- bala_text_embeddings
- bala_document_embeddings
- bala_image_embeddings

Search Configuration

Similarity threshold: > 0.7
Default limit: 5 results
Combined ranking
Source type tracking

Performance Features

Search Optimization

Efficient vector operations
Result limiting
Execution time tracking
Performance logging

Result Management

Combined result sorting
Similarity scoring
Source type tracking
Metadata preservation

Data Models

Search Result

{
    "content": str,
    "metadata": dict,
    "similarity": float,
    "source_type": str
}

Sources Used

{
    "document_sources": int,
    "image_sources": int
}

Error Handling

Query validation
Database connection errors
Search operation failures
Response generation errors
Comprehensive error logging

Future Enhancements

LLM response generation implementation
Additional content type support
Advanced context processing
Response quality metrics

Overview Carina - Search Engine