🏗️ SoRag Retrieval Architecture
Enterprise RAG System - LangGraph Workflow Visualizations
6
Main Workflows
9
Specialized Agents
59
Processing Components
22
Integration Tools
System Overview
Complete Architecture
The SoRag system is a comprehensive enterprise RAG solution built on LangGraph, featuring:
- Supervisor Orchestrator: Main controller that routes requests to appropriate workflows
- Multi-Agent System: Specialized agents for RAG, chat, and database integration
- Indexing Pipeline: Complete document processing from raw files to search-ready vectors
- Shop Integration: Database operations with 22 specialized tools
graph TB
%% Style definitions
classDef supervisor fill:#667eea,stroke:#fff,stroke-width:3px,color:#fff
classDef orchestrator fill:#764ba2,stroke:#fff,stroke-width:2px,color:#fff
classDef workflow fill:#f093fb,stroke:#333,stroke-width:2px,color:#333
classDef agent fill:#ff6b6b,stroke:#333,stroke-width:2px,color:#333
classDef node fill:#4ecdc4,stroke:#333,stroke-width:1px,color:#333
classDef startEnd fill:#4CAF50,stroke:#fff,stroke-width:2px,color:#fff
%% Main flow
START(("👤 USER QUERY")):::startEnd
END(("📋 RESPONSE")):::startEnd
%% Supervisor nodes
SUPERVISOR["🎯 SUPERVISOR - Enterprise RAG Orchestrator"]:::supervisor
QUERY_SPLITTER["📝 Query Splitter - Split complex queries"]:::node
ORCHESTRATOR["🔄 Orchestrator - Coordinate agents"]:::orchestrator
SYNTHESIZER["🔧 Synthesizer - Combine results"]:::node
%% Workflow selection
ROUTER{{"🔀 Workflow Router - chat_mode + parameters"}}:::node
%% Workflows
CHAT_WF["💬 Chat Workflow"]:::workflow
RAG_RAW_WF["🔍 RAG Raw"]:::workflow
RAG_AI_WF["🤖 RAG + AI"]:::workflow
SHOP_WF["🏪 Shop Integration"]:::workflow
%% Agents (from registry)
subgraph AGENTS["Agent Registry"]
RAG_AGENT["🔍 RAG Agent - 5-step pipeline"]:::agent
CHAT_AGENT["💬 Chat Agent - Direct LLM"]:::agent
INTEGRATION_AGENT["🏪 Integration Agent - Shop supervisor"]:::agent
end
%% Shop sub-agents
subgraph SHOP_AGENTS["Shop Agents (22 tools total)"]
SHOP_DB["🗄️ DB Agent - 11 CRUD tools"]:::agent
SHOP_ANALYSIS["📈 Analysis - 8 query tools"]:::agent
SHOP_VIZ["📊 Visualization - 3 render tools"]:::agent
end
%% Main flow
START --> SUPERVISOR
SUPERVISOR --> QUERY_SPLITTER
QUERY_SPLITTER --> ROUTER
%% Workflow selection
ROUTER -->|"chat_mode=chat"| CHAT_WF
ROUTER -->|"rag + no LLM"| RAG_RAW_WF
ROUTER -->|"rag + LLM"| RAG_AI_WF
ROUTER -->|"starts with /"| SHOP_WF
%% All workflows go through orchestrator
CHAT_WF --> ORCHESTRATOR
RAG_RAW_WF --> ORCHESTRATOR
RAG_AI_WF --> ORCHESTRATOR
SHOP_WF --> ORCHESTRATOR
%% Orchestrator uses agent registry
ORCHESTRATOR --> RAG_AGENT
ORCHESTRATOR --> CHAT_AGENT
ORCHESTRATOR --> INTEGRATION_AGENT
%% Integration agent manages shop agents
INTEGRATION_AGENT --> SHOP_DB
INTEGRATION_AGENT --> SHOP_ANALYSIS
INTEGRATION_AGENT --> SHOP_VIZ
%% Results synthesis
ORCHESTRATOR --> SYNTHESIZER
SYNTHESIZER --> END
Main Supervisor Workflow
🏗️ Complete System Architecture
View the detailed architecture by navigating through the tabs above
Each workflow has its own dedicated PNG visualization
Supervisor Workflow
Main Orchestrator
The supervisor is the main orchestrator that coordinates all workflows and agents in the system.
It handles query splitting, routing, agent coordination, and result synthesis.
Chat Workflow
Direct LLM
Simple chat mode that directly uses LLM without RAG retrieval. Used for general conversations.
Chat workflow: Start → Chat (Direct LLM) → End
RAG Raw Workflow
Retrieval Only
The RAG Raw workflow implements retrieval without LLM generation.
Pipeline: Memory Check → Retrieve → Rerank → Memory Store → Return raw results.
RAG Pipeline Architecture
RAG Pipeline (No LLM):
Start → Memory Check → Retrieve → Rerank → Memory Store → End
RAG + AI Workflow
Retrieval with LLM
The RAG + AI workflow adds LLM generation after retrieval.
Pipeline: Memory Check → Retrieve → Rerank → Memory Store → LLM Generate → Return AI response.
RAG Pipeline with LLM Generation:
Start → Memory Check → Retrieve → Rerank → Memory Store → LLM Generate → End
Shop Integration Workflow
Multi-Agent System
A supervisor-based multi-agent system with 3 specialized agents and 22 tools for database operations.
Handles CRUD operations, data analysis, and visualization.
Appointment System (Clinic / Law Firm)
Detailed Multi-Agent Architecture
Supervisor-based Multi-Agent System
🗄️ DB Agent (11 tools) | 📈 Analysis Agent (8 tools) | 📊 Visualization Agent (3 tools)