Upload a messy CSV, see exactly what's wrong with it, fix it in a few clicks, and leave with a clean file plus a plain-English report of what changed.

Real-world spreadsheets are rarely clean. They have duplicate rows, half-filled columns, inconsistent text and wrong data types. TidyBench automates the first, most repetitive hour of any data task: figuring out what's wrong before you can do anything useful with it.

Diagnose: rows, columns, missing values, duplicates and data types at a glance

Treat: fill missing values (mean, median, mode or a custom value) or drop them, remove duplicates, strip stray whitespace, and convert column types, all from the UI

Report: a before-and-after summary of everything that changed, downloadable as Markdown alongside the cleaned CSV

Quick Insights: once the data is clean, pick a preset question (top category by total, average, highest or lowest value) and get a plain-English answer

Upload your own CSV, or try the included sample_data/messy_sales.csv. It has duplicate rows, missing values, inconsistent casing and one obvious outlier built in.

Last tested in September 2026 with Python 3.13, pandas 3.0 and Streamlit 1.60.

cleaning.py has no Streamlit import, so the logic can be tested and reused on its own. Run python test_cleaning.py to check it against the sample data.

Reading and inspecting data: read_csv, .shape, .dtypes, .isnull(), .duplicated()

Cleaning: .fillna(), .dropna(), .drop_duplicates(), .astype(), .str.strip()

numpy: np.percentile for IQR-based outlier bounds

Streamlit: session state, tabs, columns, metrics, download buttons

pandas 3.0 changed the default type for text columns from object to a dedicated str type, so text columns are detected with pd.api.types.is_string_dtype(), which works on both versions

cleaning.py already has a tested find_outliers_iqr() function that isn't wired into the UI yet. That's the next feature. After that:

Outlier tab: flag rows with find_outliers_iqr, then drop or cap them

Treat inconsistent categories as one value (e.g. "NY" / "New York" / "ny")

Handle mixed date formats within a single column

Strip currency symbols and commas from numbers stored as text (e.g. "$1,200")

Clean up column names (spaces to underscores, consistent casing)

Support multi-sheet Excel uploads

Catch near-duplicates, not just exact ones

Export the summary report as a PDF

An early project (July 2026) by Rushil A. Bajpai. I now run Ru Visibility, an SEO and AI visibility (GEO) consultancy.

MIT. See LICENSE.