SourceDocka AI

Create RAG and machine-learning datasets from your phone. Capture or import images, add descriptions or annotations, export ZIP/COCO/YOLO, and sync PDFs to Google Drive for NotebookLM.

Comment

Inspiration

Creating useful datasets for AI usually means moving between a camera, a computer, file-management tools, annotation software, and cloud storage.

I wanted to make that workflow much simpler: capture or select images on a smartphone, add the information needed for RAG or machine learning, and export the result directly from the same device. That idea became SourceDocka AI.

https://sakuraforge.net/sourcedocka

What it does

SourceDocka AI is a mobile workspace for creating RAG and machine-learning datasets directly on an iPhone.

For RAG data, users can:

  • Capture photos or import images from the photo library
  • Add image names and descriptions
  • Export images and matching text together as a ZIP
  • Generate PDFs containing images and descriptions
  • Sync those PDFs to Google Drive so they can be used as sources in NotebookLM

For machine-learning datasets, users can:

  • Register images
  • Add labels and bounding-box annotations
  • Export datasets in COCO or YOLO format

The app supports Japanese and English and is designed to make AI dataset creation approachable without requiring a desktop development environment.

How I built it

SourceDocka AI was built with React Native and Expo, using TypeScript.

I used Replit throughout development to iterate on the application, UI, testing, and release preparation.

Google Drive integration uses Google OAuth and the Drive API with the drive.file scope, so the app only works with files it creates or manages.

RevenueCat powers the Premium subscription experience and free trial.

The iOS app is built and distributed with Expo EAS and is published on the Apple App Store.

Challenges I ran into

One challenge was making Google Drive authentication work reliably in a production iOS build rather than only during development.

Another challenge was generating PDFs that preserve both the original images and searchable text so they work well as knowledge sources.

I also had to design the export workflow so that RAG users and machine-learning users could share the same simple mobile experience while still supporting different output formats.

Accomplishments that I'm proud of

  • At age 63, with almost no prior programming experience, I used Replit and AI-assisted development to build and successfully release my first iOS app on the App Store during Shipaton 2026.
  • Integrated RevenueCat subscriptions
  • Built Google Drive PDF synchronization for RAG datasets
  • Added exports for RAG ZIP, COCO, and YOLO
  • Added Japanese and English localization
  • Created the entire workflow so dataset creation can be completed from a smartphone

What I learned

I learned how much work is involved beyond writing application code: App Store distribution, subscriptions, OAuth configuration, privacy disclosures, localization, production builds, and real-device testing all matter.

I also learned that AI-assisted development with Replit can make it possible to turn an idea into a real, published product even without a traditional mobile-development workflow or extensive programming experience.

Through this project, I learned that tools like Replit can lower the barrier to software development, but successfully shipping an app still requires careful testing, iteration, and attention to the complete user experience.

What's next for SourceDocka AI

Next, I want to improve how datasets are structured for AI knowledge tools, expand export options, improve annotation workflows, and continue making SourceDocka AI easier for people who want to create their own AI-ready data from a smartphone.

Built With

Share this project:

Updates

Submission history