Skip to content
JAEWON CHANGSystems Dossier
Language

Specification

Period
2025
Result
1st of 10 teams
Redaction modes
Blur, mosaic, pixelate, AI image replacement
Role
Frontend — upload, analysis results, redaction options, auth and user API integration

Context

Before sharing an image it is hard to check everything that ended up in it. Name tags, documents, screens and passers-by in the background are usually missed.

Approach

  1. Uploaded images go to an image AI server where the Gemini Vision API detects personal information and faces, and OpenCV and Pillow perform the redaction.

  2. The detections are shown to the user first, who chooses what to redact. Fully automatic processing leaves the user unsure what was removed.

  3. I built the upload, the analysis result view, the redaction option selection, the backend API integration, and sign-in with the user API.

System

Detection and redaction pathconfirmed itemsUploadSpring BootFastAPI image serverGemini VisiondetectOpenCV / Pillowuser-selected only
Fig. safelens flow

Detection is automatic; redaction covers only what the user confirmed.

An uploaded image passes through the backend to the image server, where Gemini Vision detects personal information and faces. Only the items the user confirms are redacted with OpenCV and Pillow.

Measured

Hackathon result

1 / 10

First place of ten teams.

Boundaries

  • The backend and the image AI server belonged to teammates; I owned the frontend.
  • Detection runs on an external API; no model was trained or tuned.