AuditIQ

No-code CSV and Excel data quality audits for analysts, auditors, and teams that need a shareable report before analysis, submission, or review.

AuditIQ scans a tabular dataset, flags common quality issues, writes a plain-language narrative, and generates a downloadable PDF report. It works without an AI key by using a deterministic local narrative. If you add a Gemini API key, the report narrative can be generated by Gemini.

What It Catches

• Missing values, empty rows, and high-null columns
• Duplicate rows and inferred duplicate keys
• Mixed types, invalid dates, and malformed identifiers
• Future dates and impossible numeric values
• Inconsistent casing, whitespace, and suspicious formats
• Outliers and high value concentration
• Cross-field issues such as start date after end date
• Optional schema violations for required columns, types, ranges, and formats

Demo

The GIF below shows the local Streamlit flow: upload data, run checks, review the findings, and download a PDF.

Expected Output

View a sample generated report: Titanic PDF report.

Quick Start

python -m venv .venv
.venv\Scripts\activate
pip install -r requirements.txt
streamlit run app.py

On macOS or Linux:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
streamlit run app.py

Then click Try sample data to audit samples/Titanic-Dataset.csv.

Optional Gemini Narrative

AuditIQ does not require an AI key to run. Without a key, it generates a deterministic narrative from the check results.

To enable Gemini-generated narratives, create a .env file in the project root:

GEMINI_API_KEY=your_key_here

The app uses gemini-2.5-flash-lite by default.

Schema Files

A schema file is optional. Upload a CSV or Excel file with this shape:

column_name,expected_type,min_value,max_value,allowed_formats,notes
Age,numeric,0,120,,
Embarked,text,,,,
PassengerId,identifier,,,^\d+$,Expected numeric identifier

Supported expected_type values are:

• text
• numeric
• date
• identifier

Expected Output

For the included Titanic sample, the engine loads 891 rows and 12 columns and returns findings such as:

Age: 177 blank or null values
Cabin: 687 blank or null values
Embarked: 2 blank or null values

The app then presents a quality score, issue summaries, a narrative explanation, and a downloadable PDF report.

Run Tests

pip install -r requirements-dev.txt
pytest -q

Project Structure

AuditIQ/
|-- app.py                  # Streamlit app
|-- ai/
|   `-- narrator.py         # Gemini and deterministic narratives
|-- engine/
|   |-- parser.py           # CSV/XLSX loading
|   |-- aggregator.py       # Registry-driven check runner
|   |-- schema.py           # Optional schema loader
|   `-- checks/             # 18 inference checks plus schema checks
|-- report/
|   `-- pdf_builder.py      # PDF report generation
|-- samples/
|   `-- Titanic-Dataset.csv # Sample dataset
|-- styles/
|   `-- app.css             # Streamlit styling
|-- tests/
|   `-- engine/checks/      # Unit tests for check behavior
|-- requirements.txt
`-- requirements-dev.txt

Quality Score

Each audit produces a score from 0 to 100:

score = 100 - sum((affected_rows / total_rows) * severity_weight)

Severity weights:

• bad = 1.0
• warn = 0.6
• info = 0.2

Scores above 85 usually indicate minor issues. Scores below 60 indicate data quality problems that should be addressed before downstream use.

Supported Files

• .csv
• .xlsx
• Tested with Python 3.11+
• Streamlit upload limit is configured as 200 MB in the UI copy

Limitations

• AuditIQ loads the full dataset into memory after chunked CSV reading, so very large files still need enough RAM.
• The schema format is intentionally simple and does not replace full data contracts.
• AI narratives are optional and depend on external Gemini API availability when enabled.
• Results are heuristics for triage and review; they are not a substitute for domain-specific validation.

License

MIT License. See LICENSE.