Natural Language to SQL Chatbot

NL2SQL Chatbot lets anyone query a database by just asking a question — no SQL knowledge needed. Behind the scenes, Gemini generates the query grounded in the real database schema, and a safety layer blocks anything that isn't a read-only SELECT before it can touch the data. Built with Streamlit, SQLite, and the Gemini API.

---

Project files

nl2sql-chatbot/
├── create_database.py   # generates the sample SQLite database (shop.db)
├── utils.py              # schema reader + SQL safety guard
├── nl_to_sql.py           # prompt engineering + Gemini API call
├── app.py                 # Streamlit chat UI (the actual app)
├── requirements.txt
└── README.md

Setup

Step 1 — Install dependencies

pip install -r requirements.txt

Step 2 — Get a free Gemini API key Go to https://aistudio.google.com/apikey, sign in with a Google account, and click "Create API key". It's free for moderate usage.

Step 3 — Build the sample database

python create_database.py

This creates shop.db with three tables: customers, products, orders (60 customers, 48 products, 800 orders — realistic enough to ask interesting questions).

Step 4 — Run the app

streamlit run app.py

Your browser opens automatically. Paste your Gemini API key into the sidebar, and start asking questions like:

• "Show top 10 customers by revenue"
• "Which product category sells the most?"
• "Total sales by city"
• "How many orders were placed in 2024?"

---

Skills

| SQL | Reading/writing SELECT, JOIN, GROUP BY, aggregate functions | | Prompt engineering | nl_to_sql.py — schema grounding, output constraints | | Database querying | sqlite3, pandas.read_sql_query() | | AI integration | Calling the Gemini API via google-genai SDK | | App security | Validating untrusted LLM output before executing it | | Streamlit | Chat UI, session state, sidebar, dataframes |

Live demo

nl-2sql.streamlit.app

👤 Author

Vikash Verma Aspiring Data Analyst | Excel · SQL · Power BI · Python | E-mail- vikashverma566@gmail.com

---