Data Inspector

Overview

Data Inspector is a browser-based dataset explorer and AI-assisted cleaning tool. Upload a CSV or JSON file — or click Load Sample to try it instantly with a built-in demo employee dataset — and the screen parses the file, infers a type for every column (numeric, text, date, or boolean), and gives you five tabs to explore the result plus an AI-generated cleaning recipe.

Loading Data

Three ways to get data into Data Inspector:

During parsing, each column is classified as numeric, text, date, or boolean. Boolean columns are auto-detected from values like true/false, yes/no, or 1/0. Empty cells, the literal string null, and N/A are all treated as missing values.

The Five Tabs

1. Data Table

A paginated grid of the raw rows (15 rows per page), with a row number column and one column per detected field. Missing values render as an italic null. Use the pager at the bottom to move through the dataset.

2. Column Stats

One card per column showing its detected type, missing-value count, and unique-value count. Numeric columns additionally show mean, min, and max. A red progress bar under each card visualizes the proportion of missing values for that column.

3. Correlation

A split-view layout for exploring relationships between numeric columns:

• Left panel — heatmap. An N×N Pearson correlation matrix across all numeric columns, with cell size scaling to fit the number of columns (larger cells for small datasets, smaller for wide ones) so labels never get cramped. Each cell shows the correlation coefficient (−1 to +1) on a red → gray → green scale (strong negative → no correlation → strong positive). Column headers are angled to stay readable even when names are long. Requires at least two numeric columns — with fewer, the tab shows a message instead of an empty grid.
• Right panel — detail. Click any cell (or any row in the ranked list below) to select a column pair. The panel shows a scatter plot of the two variables with a fitted trend line, a plain-language read of the relationship (e.g. "moderate positive correlation (r = 0.62) — as X increases, Y tends to increase"), and a Top Correlated Pairs list ranking every numeric pair by strength so you can jump straight to the most significant relationships.

4. Distributions

Every column is shown at once as a card in a responsive grid, so you get a full profile of the dataset in one screen instead of switching between columns one at a time:

• Numeric columns — a row of stat tiles (mean, median, min, max), a histogram (8 buckets), and a box plot showing the interquartile range and median line.
• Boolean columns — a donut/pie chart showing the split between the detected values.
• Text/date columns — a frequency bar chart of the top values, with angled labels so long category names stay readable.
• Distribution Insights card — a trailing summary card counting how many columns of each type (numeric/text/date/boolean) the dataset has, plus a few auto-generated notes flagging columns with high variance or a skewed mean-vs-median gap.

5. Data Quality

A dashboard-style quality report, not just a checklist:

• Data Health Score — a circular score (0–100%) in the top-left, derived from the share of missing cells, outlier cells, and duplicate rows relative to total data volume, labeled Good / Fair / Poor.
• KPI strip — duplicate row count, number of columns with missing values, and number of columns with detected outliers.
• Detail lists (left) — missing values per column and outliers per numeric column (IQR method, values outside [Q1 − 1.5×IQR, Q3 + 1.5×IQR]), each as a progress bar with count and percentage.
• Quick Remediation Actions (right) — one-click fixes generated from the same detection logic as the AI recipe (remove duplicates, fill missing values, remove outliers for a specific column). Clicking Apply updates the working dataset immediately, without needing to run a full AI analysis first.

AI Cleaning Recipe

The right-hand panel runs an AI analysis (triggered by the Run AI Analysis button) that inspects the loaded dataset and proposes a list of cleaning steps. Each suggested step has:

Field	Description
Kind	One of Fill Missing, Fix Format, Remove Outlier, or Standardize (used for duplicate-row removal).
Severity	High, Medium, or Low, based on how much of the dataset the issue affects.
Code Preview	A pandas-style code snippet showing how the fix would be applied, toggled with the "View Code" link on each step.
Actions	Approve or Dismiss the individual step.

Use Approve All to approve every active (non-dismissed) step at once, or approve steps individually, then click Apply Selected in the panel footer to apply only the approved steps to the working dataset. Applying steps updates the Data Table and all other tabs immediately, and a "Preview Applied" badge appears in the header.

Header Actions

Actions in the header bar follow a visual hierarchy from left to right: data-input actions, then output/utility actions, then a secondary action, then the primary Run AI Analysis button styled with the accent fill and placed at the far right.

Terminal Commands

The bottom terminal bar accepts short commands as an alternative to clicking buttons:

Command	Effect
:analyze	Run AI analysis and generate the cleaning recipe
:approve all	Approve every active recipe step
:approve <term>	Approve steps whose column, kind, or title matches the given text
:apply	Apply all currently approved steps
:reset	Revert to the original uploaded data
:export	Download the cleaned dataset as CSV
:stats	Print row/column counts and a one-line summary per column
:sample	Load the built-in sample employee dataset
:help	List all available commands

AI Assistant Panel

Data Inspector includes TotalApp's standard AI Assistant chrome, accessible from the tab on the right edge of the screen. It supports fullscreen mode, conversation history, and Save as Report to persist an AI-generated summary as a report you can revisit later.

The assistant is scoped as a data-cleaning advisor. It has access to the loaded file's name and row/column counts, per-column statistics (type, missing count, unique count, and mean/min/max for numeric columns), and the current AI cleaning recipe with each step's approval state. One-click suggestion chips cover the most common questions:

• Summarize this dataset — a plain-language read of what the data contains and its overall quality.
• Explain the cleaning recipe — a walkthrough of what each proposed step does and why.
• Which issues are most critical? — prioritization guidance across missing values, format inconsistencies, and outliers.
• Suggest a cleanup strategy — a recommended order of operations, including checks the built-in recipe generator may not have covered.

Frequently Asked Questions