Natural Language to SQL with LangChain

Query a relational database using plain English. This project connects a SQLite database to a large language model (LLM) via LangChain, allowing users to ask questions in natural language and receive answers backed by real SQL queries.

---

What This Project Does

Instead of writing SQL manually, you type a question like:

"Which country's customers spent the most?"

The LLM converts it to SQL, runs it against the database, and returns the answer:

USA — $523.06

---

Tech Stack

Component	Tool
Language	Python
LLM	Llama 3.3 70B via Groq API
LLM Framework	LangChain
Database	SQLite (Chinook)
Interface	Jupyter Notebook

---

Database

Uses the Chinook database — a sample dataset modeled after a digital music store with 11 tables including Artist, Album, Track, Customer, Invoice, and Employee.

---

Project Structure

LangChain_SQLReasoning/
│
├── C1_SQL_Chain_OpenAPI.ipynb   # Main notebook: DB connection, chain setup, query execution
├── chinook.db                   # SQLite database (included, no download needed)
├── prompts.jsonl                # Prompt templates used for LLM chain configuration
├── .env                         # API keys (not committed)
└── requirements.txt             # Dependencies

---

How It Works

Step 1 — Connect to Database LangChain's SQLDatabase wrapper connects to the Chinook SQLite database and inspects the schema automatically.

Step 2 — Initialize LLM Groq's Llama 3.3 70B is used as the reasoning engine. OpenAI GPT-3.5 is also supported by swapping one line.

Step 3 — Build the Chain SQLDatabaseChain links the LLM to the database. When a question is asked, the LLM generates a SQL query, executes it, and interprets the result.

Step 4 — Ask Questions

result = sql_db_chain.invoke("Which country's customers spent the most?")
# Returns: USA — $523.06

---

Known Limitation

SQLDatabaseChain can fail when the LLM wraps its SQL output in markdown code fences. This is a known upstream issue. The fix is migrating to create_sql_agent, which handles output parsing more robustly. This upgrade is planned in the next version of this project along with a Streamlit UI.

---

Setup

git clone https://github.com/prasad11s/text-to-sql-langchain.git
cd LangChain_SQLReasoning
pip install -r requirements.txt

Create a .env file:

GROQ_API_KEY=your_groq_api_key
OPENAI_API_KEY=your_openai_api_key  # optional

The chinook.db file is already included in this repository — no separate download needed.

---

Roadmap

• Migrate from SQLDatabaseChain to create_sql_agent for better reliability
• Build Streamlit UI for interactive natural language querying
• Deploy on Streamlit Cloud with live demo link
• Support PostgreSQL and MySQL connections

---

Author

Prasad Shimpi
MS Applied Data Science, Syracuse University
LinkedIn | GitHub