Verified repository-artifact tour

The AI data scientist that shows its work.

Data Science Agent turns questions about CSV files and databases into SQL, statistics, machine learning, visualization, claim-level evidence, and reproducible artifacts that can be inspected instead of merely trusted.

Install from PyPI Explore verified cases
Boundary: this is a static product tour built from recorded, verified DSA runs. It does not upload files, execute a new browser-side analysis, or call a live model/API.
Question
Plan
Tools
Evidence
Claim
Artifacts
Three flagship workflows

See the evidence path in real runs

CS01 · Business analytics

Sales analysis

From a stakeholder question to profiling, correlation, SQL aggregation, statistical testing, visualization, evidence capture, and a reproducible report.

1.33 s6 evidence6 tool calls5 artifacts
Inspect CS01 →
CS03 · Forecasting

Time-series workflow

A temporal workflow that preserves recorded tool failures instead of polishing them away, making recovery behavior and capability boundaries inspectable.

~1.28 s5 evidencefailures retained
Inspect CS03 →
CS08 · Machine learning

Imbalanced classification

Classification, evaluation, feature importance, evidence capture, and explicit limitations inside the same provenance-oriented workflow.

~0.11 s5 evidenceML evaluation
Inspect CS08 →
What makes it different

A claim should trace back to computation and data

Question ↓ Executed analysis ↓ Tool call / statistical result / model output ↓ Evidence item ↓ Claim ↓ report.md + experiment.json + evidence_graph.json + analysis.ipynb + reproduce.sh
InspectableImportant claims are linked to concrete analysis outputs rather than existing only as prose.
ReproducibleCompleted runs can include reports, experiment metadata, evidence graphs, notebooks, and reproduction scripts.
Failure-visibleThe verified case-study suite keeps recorded tool failures and limitations instead of deleting them from the demo record.
Run it yourself

From install to inspectable output

pip install jack-data-science-agent dsa demo
dsa analyze sales.csv \ --task "Which factors explain revenue, and are the effects statistically significant?"

Python 3.12+ is the supported baseline.

Eight verified workflows

Business analytics, churn, forecasting, marketing, finance, public statistics, data quality, and classification.

The current suite uses versioned synthetic datasets with a fixed seed and is intended as product-behavior and reproducibility evidence—not as an independent real-LLM leaderboard.

Browse all case studies Read the full Product Tour