DaTu: AI-Assisted Exploratory Data Analysis for Data Science Education
DaTu is an open-source software tool for AI-assisted exploratory data analysis that makes the initial phase of data science more reproducible and auditable by surfacing data quality issues and statistical summaries in a structured, inspectable format. Published in Software Impacts (Elsevier 2024), the tool demonstrates Fact AI Lab's commitment to building AI systems with transparent, intermediate steps that a practitioner can review and validate before proceeding.
Exploratory data analysis (EDA) is the most time-consuming and error-prone phase of most data science projects. Practitioners spend hours writing boilerplate code to inspect data shapes, distributions, missing values, and correlations - work that must be redone from scratch for each new dataset.
What DaTu Does
DaTu automates the routine EDA tasks while producing output structured for human review rather than for automated downstream consumption. The distinction matters: DaTu is not a black-box feature engineering pipeline. It is a tool that surfaces what a practitioner needs to see, then waits for their judgment before the next step.
Core capabilities:
- Automated data quality audit: missing values, duplicate rows, type inconsistencies, outlier detection
- Statistical summary with distribution plots for numeric and categorical features
- Correlation analysis with configurable significance thresholds
- Data lineage tracking: every transformation is logged with the input/output schema
Design principle: Every output is a human-readable artifact. No transformations are applied silently. The practitioner reviews the EDA report, decides what to address, and explicitly approves the resulting cleaned dataset before modeling begins.
Why This Matters
This design philosophy - explicit checkpoints, human review before transitions, no silent side effects - directly informs how Fact AI Lab structures AI pipelines for regulated industries where auditability is a compliance requirement, not a preference.
DaTu is open source and has been used in data science curricula at Simon Fraser University.