-
-
Data Sanitizer. Do it before submitting to external AI.
Inspiration
As a control systems engineer, I work with large project packages containing specifications, control narratives, sequences of operation, building drawings, electrical drawings, mechanical drawings, P&IDs etc.
These documents contain customer names, site information, addresses, network details, and other sensitive data, so we cannot just upload an entire project package directly to AI system.
I wanted a simple way for managers to prepare this material locally before giving it to Codex—without command-line tools, complicated scripts, or manually reviewing thousands of files.
That became Sanitizer.
What it does
Sanitizer is a Windows desktop application that transforms confidential project documentation into a clean, organized, AI-ready workspace.
Users create a project and submit individual files or complete folders. Sanitizer then:
- Preserves the original documents.
- Extracts text and images from supported files.
- Converts supported content into Markdown.
- Lets users define sensitive words and phrases.
- Deletes sensitive text or replaces it with a user-defined mask.
- Organizes processed documents by type.
- Creates a searchable project index.
- Generates a folder-based workspace that helps Codex locate the right documents.
- Keeps the processing workflow local to the user’s computer.
The interface is designed for managers. It provides a project tree, guided submission forms, processing status, document review, and AI-workspace export without requiring technical knowledge.
How we built it
We built Sanitizer as a native Windows desktop application designed for Microsoft Store distribution.
The application uses a structured processing pipeline:
- Import and preserve the original files.
- Identify and classify supported documents.
- Extract text, tables, metadata, and images.
- Apply user-defined deletion and masking rules.
- Verify the sanitized results.
- Convert supported content into Markdown.
- Build document indexes and relationships.
- Generate a portable workspace for Codex.
Each project remains folder-and-file based. A local database provides fast search, but it is only a rebuildable index—the project does not depend on a proprietary database to remain usable.
We used Codex with GPT-5.6 Sol throughout development to help translate the product requirements into architecture, implementation tasks, tests, debugging workflows, and a completed desktop application.
Challenges we ran into
The biggest challenge was that sanitization is more complicated than replacing words in a text file.
Sensitive information may appear in document text, filenames, folder names, metadata, PDF text layers, scanned pages, title blocks, or engineering drawings. A visual overlay is not enough if the original text remains searchable or recoverable underneath it.
Another challenge was preserving technical meaning. Removing all identifiers makes an engineering document useless. Sanitizer supports masking, allowing a customer name or site address to be replaced while keeping the surrounding sentence, drawing, and document relationships understandable.
We also had to design the workflow for non-technical users. Document conversion, OCR, indexing, and verification are complex processes, but the interface needed to present them as a simple sequence of submitting, processing, reviewing, and exporting.
Accomplishments that we're proud of
We are proud that Sanitizer is a complete working desktop application rather than a proof of concept.
It combines local document processing, configurable sanitization, content conversion, project organization, search, and Codex workspace generation in one manager-friendly workflow.
We are especially proud of the folder-based design. Users retain control of their documents, can inspect the generated material, and are not locked into the application to access their processed project.
Most importantly, Sanitizer creates a practical bridge between confidential real-world documentation and AI-assisted work.
What we learned
We learned that preparing information for AI is not simply a file-conversion problem. It is a trust problem.
Useful AI context must be safe, structured, traceable, and easy to verify. Sanitization needs an independent verification step, and users need control over what is removed, what is masked, and what is approved.
We also learned that Codex becomes much more effective when it receives a well-organized workspace instead of an unstructured collection of files. Clear indexes, document relationships, revisions, and project instructions help it find the correct source and reduce unsupported assumptions.
What's next for Sanitizer
Next, we plan to:
- Expand support for additional engineering and business-document formats.
- Improve sanitization of scanned drawings and title blocks.
- Add reusable sanitization policies for organizations and customers.
- Improve automatic detection of document types, revisions, and equipment tags.
- Build richer relationships between specifications, narratives, sequences, and drawings.
- Add stronger validation and reporting for approved AI workspaces.
- Prepare Sanitizer for release through the Microsoft Store.
The long-term goal is to make confidential project knowledge safely usable with AI—while keeping organizations in control of their information.
Log in or sign up for Devpost to join the conversation.