🤖 Agentic AI - Data Agent

An intelligent, multi-agent system built with LangGraph that processes natural language queries by routing them to specialized agents for SQL database analytics .

---

🏗️ System Architecture

                   ┌─────────────────────────┐
                   │   Data Agent (Router)   │
                   └────────────┬────────────┘
                                │
                                ▼
                      ┌──────────────────┐                
                      │  SQL Analyst     │
                      └─────────┬────────┘                
                                │
                                ├─► Context & Schema Gathering      
                                ├─► SQL Generation & Safety Check   
                                └─► Query Execution & Answers       

---

✨ Core Features

• Intelligent Intent Routing: Automatically classifies natural language user input into SQL.
• SQL Analyst Agent: Converts natural language to safe PostgreSQL queries, performs strict schema validation, and blocks destructive operations (DROP, DELETE, UPDATE).
• Dynamic Model Tiering: Routes queries across fast/cost-effective LLMs or premium models (e.g., Claude 3 Opus) based on task complexity.

---

🚀 Quickstart

1. Prerequisites & Setup

• Python: 3.12+
• Database: PostgreSQL instance running locally or remotely.

# Clone repository and navigate inside
cd Data_Agent

# Create and activate virtual environment
python -m venv .venv
source .venv/bin/activate  # On Windows: .\.venv\Scripts\Activate.ps1

# Install dependencies
pip install -r requirements.txt

2. Environment Configuration

• Create a .env file in the root directory:

    # API Keys
  ANTHROPIC_API_KEY=your_claude_api_key
  OPENAI_API_KEY=your_openai_api_key
  
  # Database Connection
  host=localhost
  port=5432
  user=postgres
  password=your_password
  database=data_agent_db
  
  # Model Selection
  LLM_MODEL_LOW=gpt-3.5-turbo
  LLM_MODEL_MEDIUM=gpt-4-turbo
  LLM_MODEL_HIGH=claude-3-opus

💻 Usage

Run Main Agent

from agents.data_agent import data_agent
from langchain_core.messages import HumanMessage

# Example 1: Database Query (SQL Agent)
sql_response = data_agent.invoke({
    "messages": [HumanMessage(content="Show me the top 5 users with the highest ratings")],
    "route_response": ""
})

Run via Command Line

# Execute full workflow
python main.py

# Run individual sub-agents directly
python agents/sql_analyst.py

📁 Repository Structure

Data_Agent/
├── agents/             # Sub-agent implementations (router, sql)
├── Models/             # Pydantic schemas and state management
├── utils/              # PostgreSQL drivers, & dynamic LLM selectors
├── data/               # Input/output workspace (extract/ and transform/)
├── feed_db.py          # Database initialization script
└── main.py             # Main entry point

🛡️ Security Features

SQL Guardrails: Queries undergo strict safety parsing prior to database execution to prevent data mutation or deletion.

Sandboxed Code Execution: Standardized input validation and restricted environment limits for runtime Pandas operations.