An AI-powered Business Intelligence agent that transforms natural language questions into data-driven insights using SQL, Python, LLM reasoning, and BI analytics workflows.
"Why did sales drop last month?" — just ask, and the agent answers.
Business users often rely on analysts to write SQL queries and prepare reports, creating delays in accessing insights.
This project demonstrates how AI can enable self-service analytics by allowing stakeholders to ask questions in natural language and receive data-driven answers instantly — without requiring SQL knowledge.
The solution combines natural language processing, SQL generation, data analysis, and automated insight generation into an interactive BI assistant.
Instead of manually writing SQL queries and analysing spreadsheets, stakeholders can simply type a business question in plain English and receive:
- Auto-generated SQL query
- Data pulled live from the database
- AI-written business insight (Summary → Key Findings → Root Cause → Recommendation)
- Interactive visualisation
- Suggested follow-up questions
Example questions you can ask:
- "What are the top 5 product categories by revenue?"
- "Which state has the most customers?"
- "What is the most popular payment method?"
- "What is the monthly revenue trend in 2018?"
- "Which product category has the highest cancellation rate?"
User Question (Natural Language)
↓
Streamlit UI (app.py)
↓
AI Agent (agent.py)
↓
┌─────────────────────────────────┐
│ LLM #1: Generate SQL │ ← Groq (Llama 3.1)
│ SQLite: Run Query │ ← Olist Database
│ Pandas: Analyse Results │ ← Data Summary
│ LLM #2: Generate Insight │ ← Groq (Llama 3.1)
└─────────────────────────────────┘
↓
Structured Output:
Summary · Key Findings · Root Cause · Recommendation
| Layer | Technology |
|---|---|
| Language | Python 3.x |
| AI / LLM | Groq API (Llama 3.1 8B) |
| Agent Framework | LangChain |
| Database | SQLite (Olist Dataset) |
| Data Analysis | Pandas |
| Visualisation | Plotly |
| UI | Streamlit |
ai-bi-agent-olist/
│
├── app.py # Streamlit UI
├── agent.py # AI agent pipeline
│
├── tools/
│ ├── sql_tool.py # SQL query execution
│ └── analysis_tool.py # Pandas data analysis
│
├── prompts/
│ └── system_prompt.txt # LLM behaviour rules
│
├── utils/
│ └── db.py # Database connection
│
├── database/
│ └── olist.db # SQLite database (tracked via Git LFS)
│
├── screenshots/
│ ├── demo_overview.png
│ ├── demo_insights.png
│ └── demo_chart.png
│
├── requirements.txt
└── README.md
git clone https://github.com/joyceleehy/AI-BI-Agent-Olist.git
cd AI-BI-Agent-Olistpip install -r requirements.txtCreate a .env file in the project root:
GROQ_API_KEY=your_groq_api_key_here
Get a free API key at console.groq.com
streamlit run app.py| Feature | Description |
|---|---|
| Natural Language to SQL | LLM automatically writes SQL from plain English |
| Real Data | Queries live against Olist database (100k+ orders) |
| KPI Cards | Instant at-a-glance metrics (rows, total, average) |
| Structured Insights | Summary → Key Findings → Root Cause → Recommendation |
| Smart Visualisation | Auto-selects bar, line, or pie chart based on data |
| Follow-Up Questions | Suggests relevant next questions dynamically |
| Analysis History | Sidebar tracks previous questions in the session |
| Raw Data Preview | Expandable table showing up to 20 rows |
Olist Brazilian E-Commerce — 100k orders from 2016–2018 across Brazil.
| Table | Description |
|---|---|
| orders | Order status and timestamps |
| customers | Customer location data |
| order_items | Products and prices per order |
| payments | Payment method and value |
| products | Product details and category |
| category_translation | Portuguese → English category names |
- User types a question in plain English
- LLM #1 (Groq) reads the database schema and writes a SQL query
- SQLite runs the query and returns raw data
- Pandas summarises the data into plain text
- LLM #2 (Groq) reads the summary and writes a structured business insight
- Streamlit displays everything — SQL, data, chart, and insight
The key insight: no SQL knowledge required from the user. The agent handles everything automatically.
Built by Joyce Lee — Data & BI Analyst in HR Analytics and Business Intelligence.
This project demonstrates:
- AI agent architecture and pipeline design
- Natural language to SQL generation
- LLM prompt engineering
- End-to-end data pipeline thinking
- Streamlit application development
This project is open source and available under the MIT License.