orion-data-analyst 1.1.4

AI-powered BigQuery data analysis agent with natural language interface

🌟 Orion - AI-Powered Data Analysis Agent

An intelligent data analysis agent that transforms natural language questions into SQL queries, executes them on BigQuery, performs statistical analysis, and generates actionable business insights.

🔗 GitHub: https://github.com/gavrielhan/orion-data-analyst
📦 PyPI: https://pypi.org/project/orion-data-analyst/

---

✨ What is Orion?

Orion is your AI business analyst that:

• Understands natural language - Ask questions in plain English
• Generates smart SQL - Powered by Google Gemini AI
• Analyzes data automatically - Statistical analysis, trends, segmentation
• Provides insights - Actionable recommendations with business context
• Creates visualizations - Charts saved automatically
• Self-heals errors - Automatically fixes and retries failed queries
• Remembers conversations - Handles follow-up questions with context

Built with LangGraph for modular AI reasoning and Google BigQuery for data warehousing.

---

🚀 Quick Start

Installation

Option 1: Install from PyPI (Recommended)

pip install orion-data-analyst

Option 2: Install from Source

git clone https://github.com/gavrielhan/orion-data-analyst.git
cd orion-data-analyst
pip install -e .

Setup

1. Get API Keys (see GETTING_KEYS.md):

   • Google Cloud Project ID
   • Google Cloud service account JSON key
   • Gemini API key from Google AI Studio

2. Configure .env file:

# Copy example
cp .env.example .env

# Edit with your credentials
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
GEMINI_API_KEY=your-gemini-api-key

3. Run Orion:

orion

---

💡 Usage Examples

Basic Queries

You: show me top 10 products by revenue
Orion: [Generates SQL, executes, analyzes, and displays ranked results]

You: what are the sales trends for the last 6 months?
Orion: [Creates time-series analysis with month-over-month growth]

You: segment customers by purchase behavior
Orion: [Performs customer segmentation and analysis]

Follow-up Questions

You: show top customers
Orion: [Displays ranked customer list]

You: show the same for the last quarter
Orion: [Uses conversation context to apply date filter]

You: break that down by region
Orion: [Further segments the previous results]

Visualizations & Exports

You: create a bar chart of sales by category
Orion: [Generates chart and saves to ~/orion_results/]

You: save this as csv
Orion: [Exports results to ~/orion_results/results_TIMESTAMP.csv]

Meta-Questions (Instant Responses)

You: what can you do?
Orion: [Explains capabilities without querying database]

You: which datasets can you query?
Orion: [Lists available tables and schemas]

---

🏗️ Architecture

Orion uses a modular node-based architecture powered by LangGraph:

High-Level Architecture

Detailed Graph Schema

See ARCHITECTURE.md for detailed component descriptions.

---

🎯 Key Features

🤖 Intelligent SQL Generation

• Natural language to SQL using Google Gemini
• Automatic schema context injection
• Self-healing with error feedback loops (max 3 retries)
• Handles complex JOINs across multiple tables

🛡️ Safety & Validation

• Blocks malicious queries (DROP, DELETE, etc.)
• BigQuery cost estimation before execution
• Query syntax validation with dry-run
• Row limits to prevent runaway queries
• Human-in-the-loop approval for expensive operations

📊 Advanced Analytics

• Ranking: Top N analysis with contribution %
• Trends: Time-series with growth rates
• Segmentation: Group-by analysis
• RFM Analysis: Customer segmentation (Recency, Frequency, Monetary)
• Anomaly Detection: Outlier identification
• Comparative Analysis: Period-over-period comparison

💬 Conversation Memory

• Remembers last 5 interactions
• Context-aware follow-up questions
• Session save/load for long conversations
• Automatic context pruning for token efficiency

📈 Visualizations

• Chart Types: Bar, Line, Pie, Scatter, Box, Candle
• Auto-saved to ~/orion_results/ (configurable)
• Smart chart type selection based on data
• CSV export for further analysis

⚡ Performance Optimizations

• Query Caching: Instant responses for repeated queries (1-hour TTL)
• Schema Caching: Reduces API calls to BigQuery metadata
• Rate Limiting: Token bucket algorithm for Gemini API
• Streaming: Large result handling

🎨 Polished UX

• Colored terminal output with formatted text
• Progress indicators at each step
• Helpful error messages with action links
• Startup validation with setup guidance

---

🗂️ Project Structure

orion-data-analyst/
├── assets/                        # Images and diagrams
│   ├── orion_face.png            # Main interface screenshot
│   ├── high_level_schema.png     # High-level architecture diagram
│   └── graph_schema.png          # Detailed graph flow diagram
├── src/
│   ├── __init__.py
│   ├── cli.py                    # CLI interface with session management
│   ├── config.py                 # Configuration loader (.env)
│   ├── agent/
│   │   ├── __init__.py
│   │   ├── graph.py              # LangGraph workflow orchestration
│   │   ├── nodes.py              # All 10 agent nodes
│   │   └── state.py              # Centralized AgentState (TypedDict)
│   └── utils/
│       ├── __init__.py
│       ├── cache.py              # Query result caching
│       ├── formatter.py          # ANSI terminal formatting
│       ├── rate_limiter.py       # API rate limiting
│       ├── schema_fetcher.py     # BigQuery schema utilities
│       └── visualizer.py         # Chart generation (matplotlib/seaborn)
├── tests/                         # Test suite
│   ├── __init__.py
│   ├── test_nodes.py             # Node unit tests
│   └── test_graph.py             # Graph integration tests
├── .env.example                  # Configuration template
├── requirements.txt              # Dependencies
├── setup.py                      # PyPI packaging
├── pyproject.toml                # Modern Python packaging
├── install.sh                    # One-line installer
├── ARCHITECTURE.md               # Detailed architecture docs
├── GETTING_KEYS.md               # API key setup guide
└── README.md                     # This file

---

⚙️ Configuration

All configuration via .env file:

# REQUIRED
GOOGLE_CLOUD_PROJECT=your-project-id
GOOGLE_APPLICATION_CREDENTIALS=/path/to/service-account.json
GEMINI_API_KEY=your-gemini-api-key

# OPTIONAL
GEMINI_MODEL=gemini-2.0-flash-exp              # Choose Gemini model
ORION_OUTPUT_DIR=~/orion_results               # Results directory
BIGQUERY_DATASET=bigquery-public-data.thelook_ecommerce
MAX_QUERY_ROWS=10000                           # Row limit
QUERY_TIMEOUT=300                              # Timeout (seconds)

---

📊 Dataset

Uses Google BigQuery's public e-commerce dataset:

• Dataset: bigquery-public-data.thelook_ecommerce
• Tables: orders, order_items, products, users
• Schema: Automatically loaded with column descriptions

---

🔧 Development

Run from Source

git clone https://github.com/gavrielhan/orion-data-analyst.git
cd orion-data-analyst
pip install -e .
orion

---

📝 Commands

In the Orion CLI:

• exit / quit / q - Exit Orion
• save session - Save conversation history
• load session [path] - Load previous session
• clear cache - Clear query cache

---

🛠️ Technology Stack

Component	Technology
AI Orchestration	LangGraph
LLM Integration	LangChain
AI Model	Google Gemini 2.0 Flash
Data Warehouse	Google BigQuery
Data Processing	pandas
Visualization	matplotlib, seaborn
State Management	TypedDict (Python)
Configuration	python-dotenv
Packaging	setuptools, PyPI

---

📜 License

MIT License - see LICENSE file for details.

---

🤝 Contributing

Contributions welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

---

🙏 Acknowledgments

• Built with LangGraph by LangChain
• Powered by Google Gemini
• Data from BigQuery Public Datasets

---

Made with ❤️ by Gavriel Hannuna