A data science team that fits inside a requirements.txt

What it does

The library bundles specialized agents for data science grunt work—loading, wrangling, cleaning, visualization, feature engineering, and modeling—into reusable Python blocks. Its flagship face is AI Pipeline Studio, a Streamlit app that turns those steps into a visual workflow with lineage tracking, multi-dataset support, and reproducible scripts. Under the hood it can talk to OpenAI or Ollama, and it integrates with H2O and MLflow when you need heavier lifting.

The interesting bit

Most agent demos stop at a chat window; this one wraps a full visual pipeline editor around the agents, complete with storage-footprint controls and metadata-only saves so you don’t accidentally commit gigabytes of intermediate data. A Supervisor Agent keeps the specialists in line, which is either solid architecture or a subtle commentary on actual data science teams.

Key highlights

• Modular agents for SQL, EDA, data cleaning, and feature engineering
• Visual pipeline editor mixing manual and AI-generated steps
• Multi-dataset handling, merge workflows, and project rehydration
• Local-LLM support via Ollama; runs offline if you want
• H2O modeling and MLflow tracking integrations

Caveats

• Explicitly beta; breaking changes are expected until the 0.1.0 release

Verdict

Worth exploring if you want to hand off repetitive data science tasks to LLMs but still need a visual audit trail. Look elsewhere if you require a mature, stable framework today; the README itself warns that it’s still settling into shape.