System Architecture

The system is designed as a state-driven graph. Instead of a linear script, the agent has the "agency" to decide its own path based on the database's feedback.

1. Intent Classification: A guardrail node that filters out non-SQL or off-topic queries.
2. Semantic Retrieval: Queries a Qdrant Vector Store to inject relevant column metadata into the LLM context, preventing hallucinations.
3. The ReAct Loop:
   • Agent: Generates SQL queries using openai/gpt-oss-20b (via Groq).
   • Tool: Executes queries against the SQLite database using SQLAlchemy.
   • Feedback: If a query fails, the agent receives the error and self-corrects in the next loop.
4. Streamlit UI: A ChatGPT-style interface that displays the agent's thought process and status updates in real-time.

Tech Stack

Language : Python 3.10+

LLM Orchestration : LangGraph & LangChain

LLM : Groq (openai/gpt-oss-20b)

Vector DB : Qdrant (In-memory)

Database : SQLite

Embeddings : HuggingFace (all-mpnet-base-v2)

Table Details

name: The full name of the employee which is unique

designation: The job title or role

department: The department name (AI, HR, Sales)

salary: The numeric value representing annual pay

id: The unique primary key

1. Prerequisites

Ensure you have Python 3.10+ installed. You will also need a Groq API Key.

-- Configuration

Create a .env file in the root directory:

Code snippet: GROQ_API_KEY=your_groq_api_key_here

2. Installation

# Clone the repository
git clone "[git repo](https://github.com/Hrithik-coder/SQL_AGENT.git)"

cd agentic-sql-analyst

# Install dependencies
pip install -r requirements.txt

# run employee_table.py for the table creation
python employee_table.py

first add some employee data by querying the agent then perform the various actions

# run the streamlit application
streamlit run app.py