Agentic Data Scientist

Detailed Introduction

Agentic Data Scientist is an adaptive multi-agent framework for data science that separates planning from execution and continuously validates results. It decomposes complex analyses into executable stages using an iterative plan–execute–validate–reflect loop. The project integrates Google ADK and Claude Agent SDK, supports MCP-based tool access and scientific skills, and offers both full orchestrated and lightweight execution modes.

Main Features

• Agent orchestration: produces staged plans and runs them with continuous validation.
• Continuous validation and self-correction: each stage has success criteria and can trigger replanning when needed.
• Rich tools & skills: integrates Claude Scientific Skills, web fetch and file tools to enhance capabilities.
• Usability & deployment: distributes as a Python package, CLI, and uvx runnable artifact for easy local or private deployment.

Use Cases

Suitable for complex analyses and report generation (e.g., differential expression, churn modeling), automated experiment pipelines, scientific data processing, and any enterprise workloads that require reproducibility and rigorous review. The framework preserves working directories and produces audit-friendly execution traces.

Technical Features

• Multi-phase workflow: planning, review, parsing, stage orchestration, execution and reflection components coordinate iteratively.
• Event compression & context management: compresses and summarizes events to manage long-running context windows.
• Extensible skills system: auto-loads Claude scientific skills and supports custom tool integrations.
• Open-source license: MIT-licensed for research and enterprise customization.