AgentADA: An Advanced Data Analytics and Evaluation Framework

Agent ADA is a comprehensive evaluation and data analytics framework focused on insights generation and skills assessment.

Features

• Insights Generation: Generate and evaluate data-driven insights
• Skills Assessment: Evaluate analytical capabilities
• Batch Processing: Support for processing multiple datasets
• Interactive Visualization: Using Gradio for question-answer-plot interactions
• LLM Integration: Advanced language model evaluation capabilities
• Automated Goal Generation: Smart goal setting and question generation
• Data Encoding Management: Specialized encoding conversion tools
• Visualization Tools: JSON-based visualization capabilities

Installation

# Set Python Path in the project root
export PYTHONPATH=$(pwd):$PYTHONPATH

Usage

1. Insights Evaluation

python main.py -e insights

2. Skills Evaluation

python main.py -e skills

3. Data Generation

python main_gen.py

Dataset Resources

The framework includes extensive datasets organized in batches. Access the Kaggle Bench CSV files:

• Batch 1
• Batch 2
• Batch 3
• Batch 4
• Batch 5
• Batch 6
• Batch 7

Project Structure

agent-ada/
├── data/           # Dataset and resource files
├── prompts/        # Prompt templates and configurations
├── scripts/        # Utility scripts
├── src/           # Source code
└── main.py        # Main execution script

Key Components

1. LLM Evaluation (LLM_only_EVAL.py): Evaluate Language Model performance
2. Batch Generation (batch_generator.py): Generate and process data batches
3. Question-Answer-Plot Interface (gradio_QuesAnsPlot.py): Interactive visualization
4. Goal Generation (main_gen_goal.py): Generate analytical goals
5. Question Generation (main_gen_questions.py): Generate analytical questions

##Citation

@misc{abaskohi2025agentadaskilladaptivedataanalytics,
      title={AgentAda: Skill-Adaptive Data Analytics for Tailored Insight Discovery}, 
      author={Amirhossein Abaskohi and Amrutha Varshini Ramesh and Shailesh Nanisetty and Chirag Goel and David Vazquez and Christopher Pal and Spandana Gella and Giuseppe Carenini and Issam H. Laradji},
      year={2025},
      eprint={2504.07421},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2504.07421}, 
}

🤝 Contributing

Please check the outstanding issues and feel free to open a pull request. For more information, please check out the contributing guidelines and issue template.

Acknowledgments

Developed and maintained by ServiceNow. Join our community to collaborate on advancing data analytics and evaluation frameworks.

Thank you!