The problem
Datasets may contain names, email addresses, identification numbers, locations, and other sensitive information. Manual anonymisation can be inconsistent and may not clearly show what was changed or whether risks still remain.
Back to Selected Work
Detailed project walkthrough
DataShield
A local-first, privacy-aware anonymization workflow app that helps users identify sensitive information, review anonymisation suggestions, apply approved changes, and check whether privacy risks may still remain.
Deployed aplication prototypeLocal-first data processingTested with synthetic data
What the application does
DataShield reviews the uploaded data, identifies possible privacy risks, recommends anonymisation actions, lets the user review the plan, applies approved changes, and checks the result.
Why it may be useful
It helps prepare data that is safer to analyse or share while keeping privacy decisions reviewable and producing repeatable reports and export files.
System design
Privacy decisions remain reviewable
The application separates recommendations, user decisions, applied changes, and final validation instead of treating anonymisation as one automatic step.
System thinking
- Sensitive columns are detected and classified before anonymisation actions are selected.
- Recommendations are separated from execution so the user can review the plan before changing the dataset.
- Privacy risk is checked again after the transformations have been applied.
- The application creates audit-ready files showing the decisions, changes, and final results.
Workflow
Dataset uploadData profilingPII detectionRisk classificationAnonymisation planTransformationRisk validationReports
Application walkthrough
From dataset upload to protected outputs
The application guides the user through each privacy decision instead of applying anonymisation as one automatic and unexplained action.
01
Upload the dataset and define its purpose
The user uploads a dataset and selects the business context. This gives the application information needed to prepare suitable privacy recommendations.
02
Identify and classify sensitive information
DataShield shows which columns were flagged and separates direct identifiers, sensitive information, and indirect identifiers.
03
Review the suggested anonymisation plan
The application recommends actions based on the detected risks and selected policy context. The user can review these suggestions before applying any changes.
04
Download the protected dataset and audit files
The completed run produces a protected dataset, transformation logs, validation results, metadata, and audit reports.
Privacy workflow
How the application works
The workflow guides the user from the original dataset to reviewed anonymisation changes and final export files.
- 01
Upload and profile the dataset
The user uploads a dataset. The application reviews its structure, column types, missing values, and other information needed for the privacy assessment.
- 02
Identify sensitive information
The application checks columns for names, email addresses, identification numbers, locations, and other information that may identify a person.
- 03
Classify privacy risk
Columns are grouped as direct identifiers, indirect identifiers, sensitive attributes, or general data. This helps explain why each field may need protection.
- 04
Review the anonymisation plan
DataShield suggests suitable anonymisation actions. The user can review and change the plan before any transformation is applied.
- 05
Apply and validate the changes
The approved transformations are applied, and the application checks whether important privacy risks may still remain.
- 06
Create reports and export files
The application creates the anonymised dataset together with reports showing what was detected, what was changed, and the result of the final privacy review.
Validation and outputs
The process does not stop after anonymisation
DataShield checks the transformed dataset and creates files that help explain what happened during the run.
Residual-risk review
The transformed dataset is checked again to identify privacy risks that may still remain.
Anonymised dataset
The user receives the transformed dataset as a separate output instead of replacing the original file.
Audit report
The report records detected risks, selected actions, applied changes, and the final validation result.
My contribution
What I worked on
- Designed the complete privacy review workflow
- Created rules for identifying and classifying sensitive columns
- Built the anonymisation planning and review stage
- Implemented data transformation handling
- Added privacy-risk checks after anonymisation
- Created export files and audit reports
- Built the Streamlit interface
- Added automated tests and project documentation
Technology used
Main tools
- Python
- Streamlit
- Pandas
- Privacy classification rules
- Anonymisation methods
- Automated tests
- Git and GitHub
- Streamlit Community Cloud
Current limitation
DataShield is a portfolio prototype tested with synthetic datasets. It supports a structured privacy review, but it does not replace a legal privacy assessment or a specialist review for real organisational data.
* Free hosted demo — it may take a moment to wake if inactive.