Natural Language → SQL Agent

Ask business questions in plain English. Get SQL + answers instantly.

Built with Python, Anthropic Claude API, SQLite, and Streamlit.

---

What it does

• Type any business question in plain English
• Claude generates the correct SQL query
• Query runs against a real sales database (500 customers, 20 products, 3,000 orders)
• Results display as a table + automatic bar chart
• Plain-language explanation of what the query found

Quick Start

1. Install dependencies

pip install anthropic streamlit pandas

2. Set your Anthropic API key

export ANTHROPIC_API_KEY="your-key-here"

3. Create the database

python create_db.py

4. Run the app

streamlit run app.py

---

Example questions you can ask

• "What are the top 5 products by total revenue?"
• "Which customer segment generates the most revenue?"
• "Show me the 10 customers with the highest lifetime value"
• "How has monthly revenue trended over time?"
• "What percentage of orders were cancelled last year?"
• "Which regions have the highest average order value?"

Live Demo

Watch it in action: https://www.loom.com/share/8566a0d4d3074f0daedfb4175169e786

Database schema

Table	Key columns
customers	customer_id, name, segment, region, signup_date, is_active
products	product_id, product_name, category, unit_price
orders	order_id, customer_id, order_date, status, discount_pct
order_items	item_id, order_id, product_id, quantity, unit_price, line_total

---

How it works

1. User types a natural language question
2. Claude receives the database schema + the question
3. Claude returns structured JSON with SQL + plain-language explanation
4. App runs the SQL against SQLite and displays results
5. If the query fails, the app automatically retries with the error message

---

Tech stack

• LLM: Anthropic Claude (claude-sonnet-4-20250514)
• Frontend: Streamlit
• Database: SQLite via Python sqlite3
• Data: pandas

---

Built as part of Tasmia Afroze's AI/analytics portfolio — tasmiaafroze.com