Orchestrating Multi Step Financial SQL Agent Using Langgraph, LangChain, FAISS Memory, and SQLite

🧠 Project Overview

This repository demonstrates a production-ready architecture for a multi-step Financial SQL Agent using cutting-edge LLM tooling. It leverages a combination of:

• LangGraph: For orchestrating multi-step agent logic.
• LangChain: For creating SQL agents and handling database toolkits.
• OpenAI LLMs: Used via ChatOpenAI to interpret user intent and generate SQL queries.
• FAISS-based Vector Store: Stores and retrieves semantic memory to provide historical context.
• SQLite: A lightweight SQL database backend with sample financial data.

---

🔍 What This Agent Can Do

• Accept natural language questions about financial data (e.g., "Show me total expenses in January").
• Retrieve contextual memory from a vector database to enrich SQL generation.
• Use LangGraph to orchestrate query steps.
• Translate the result into a clear and useful output.

---

🗂️ Project Structure

financial-sql-agent/
├── agents/
│   ├── query_agent.py       # Builds the SQL agent using LangChain + OpenAI + memory
│   └── orchestrator.py      # Orchestrates query handling
│
├── data/
│   ├── init_db.py           # Initializes SQLite with sample financial transactions
│   └── sample_financial_data.db
│
├── memory/
│   └── vector_store.py      # FAISS vector memory using OpenAI embeddings
│
├── prompts/
│   └── financial_prompt.txt # System prompt that guides the agent
│
├── langgraph_app/
│   └── langgraph_flow.py    # LangGraph setup for execution flow
│
├── main.py                 # CLI app entry point
├── requirements.txt        # Dependencies
└── README.md               # Project documentation

---

🚀 How to Run This Project

1. Clone the Repo

git clone https://github.com/your-username/financial-sql-agent.git
cd financial-sql-agent

2. Create a .env File

Create a .env file in the root folder:

OPENAI_API_KEY=your_openai_api_key

3. Install Dependencies

pip install -r requirements.txt

4. Initialize the SQLite DB

python data/init_db.py

This will create sample_financial_data.db with mock transactions.

5. Run the App

python main.py

Then enter questions like:

• "What were my grocery expenses in January?"
• "Total salary received in 2024?"

---

⚙️ Key Components & Techniques

✅ LangGraph (Multi-Step Workflow)

LangGraph allows stateful orchestration between agents. Here it's used to:

• Trigger the query agent
• Manage memory and execution flow

✅ LangChain SQL Agent

• Automatically translates user queries into optimized SQL
• Uses SQLDatabaseToolkit to explore schema and generate SQL
• Uses AgentType.ZERO_SHOT_REACT_DESCRIPTION for efficiency

✅ Memory with FAISS Vector Store

• Stores and retrieves past context semantically
• OpenAIEmbeddings + FAISS power fast and relevant memory lookup

✅ OpenAI Chat Model

• Uses ChatOpenAI(temperature=0) for deterministic query generation
• Context is injected into the prompt dynamically

✅ SQLite Backend

• Lightweight and easy to explore
• Schema:

CREATE TABLE transactions (
  id INTEGER PRIMARY KEY,
  date TEXT,
  amount REAL,
  category TEXT,
  account TEXT
);

---

📌 Future Improvements

• Add multi-table schema support
• Integrate LangSmith for debugging
• Expose via REST API or Streamlit UI
• Add user authentication and persistent memory

---

🧑‍💻 Author

Crafted by Adad — Machine Learning Engineer.

Need help deploying this project or scaling it with real data? Feel free to ask!