Capabilities

Everything you need to understand a dataset

From the first upload to a downloadable executive report — here is what runs under the hood, all grounded in deterministic computation.

Automatic profiling

Row/column counts, dtype inference, semantic column mapping, and a per-column quality profile the moment you upload.

Data quality scan

Missing-value severity, duplicate detection, constant and high-cardinality columns, and IQR-based outlier flags.

Business metrics & EDA

Revenue, margins, top categories/products, trends, and seasonality — all computed deterministically in Pandas.

Trends & correlations

Period-over-period movement, volatility, anomaly points, and a Pearson correlation matrix with strong-pair detection.

Prediction & model evaluation

An optional Ridge / RandomForest model with a real held-out evaluation — R², RMSE, MAE, or accuracy and F1.

Explainable AI (SHAP)

Global feature importance and plain-English explanations so a prediction is never a black box.

Data-leakage analysis

Flags suspicious near-perfect correlations and target leakage, with financial scale-effects handled correctly.

Compact, de-duplicated report

A 1–2 page HTML/PDF with KPIs, charts, recommendations, and XAI — every fact appears exactly once.

What's inside the report

Concise by design — and free of repeated content.

Dataset snapshot

Filename, shape, detected type, and the important columns.

Key metrics

The headline KPIs — surfaced once, never repeated downstream.

Data quality

Quality score, missing values, duplicates, and the top warning.

Important insights

The highest-signal findings, ranked and de-duplicated.

Charts

Up to two auto-selected visualisations with takeaways.

Model performance evaluation

R²/RMSE/MAE or accuracy/F1 with a plain-English read and leakage note.

Recommendations

At most three concrete, prioritised next steps.

Explainable AI

SHAP feature importance — always the final section.

Works fully offline in Mock mode with no API keys — the LLM is optional and only refines wording.

Try it on your own data