Summary
Privacy Guard for AI
I designed and engineered a Chrome extension as a solo project to prevent AI oversharing. Working across UX and development, I built real-time privacy detection that runs 100% locally. To achieve this, I coded a lightweight system that catches passwords, SSNs, and secrets before they reach AI—alerting users without disrupting their workflow.
Skills
UI/UX Design
Prototyping
Engineering
Roles
Product Designer
UX Engineer
Developer
Collaborators
Me
Myself
and i
Duration
14 Days
Context
Wait... did I just share THAT with AI?
Major companies, government agencies, and millions of users have already paid the price for AI oversharing.
News screenshots about AI data leakage
Problem
Users frequently overshare sensitive information with AI without realizing it.
In the flow of chatting with AI, sharing sensitive details feels natural and harmless.
Data Leaks
Personal details, work secrets, financial info slip into conversations
Trust Gap
High user trust in AI creates blind spots for data privacy
No Protection
No built-in privacy protection in popular AI tools
Missing Privacy Tools
No convenient way to de-identify sensitive information before sharing
User interview
Convenience leads to unintentional privacy risks.
Whether it’s work notes, personal info, or even IDs, they only realize it might be risky after hitting enter.
Hover to Zoom In
Users want AI help that feels seamless and invisible. They value being fast, simple, and uninterrupted while staying “in flow.” At the same time, this desire for smooth interaction creates blind spots where oversharing occurs without awareness.
Design Principles
User insights guided us in defining four key design principles.
Instead of blocking users outright, small contextual cues can make them aware while maintaining trust and productivity.
Key Functions
Real-time Privacy Nudges for AI Chats
Based on the design principles, we came up with four key functions.
Real-time Detection
Instantly alerts you when sensitive data is detected as you type
Built-in Protection Rules
Automatically identifies multiple data types without configuration
Custom Rules
Create your own privacy rules without writing code
100% Local Processing
Your data never leaves your device. Complete privacy guaranteed.
Customer Journey Map
How Privacy Guard AI Should Transform User Experience
This journey map represents the design vision for how Privacy Guard AI should integrate into users' natural workflow.
Hover to Zoom In
The temporary confidence dip during Privacy Warning stage is crucial. It represents the moment users gain true privacy awareness. The final high confidence level is more valuable than initial confidence because it's based on understanding, not ignorance.
Ideation & Concept Development
Exploring Notification Patterns
I explored multiple approaches to privacy notifications, balancing user awareness with workflow preservation.
D: Center-screen modal (too proactive, would heavily disrupt user flow)
Prototyped
B: Enhanced popup with actionable buttons and clear "sensitive data" labeling
Prototyped
C: Corner-positioned notification for less intrusive experience
A: Minimal popup with basic privacy warning (doesn’t have enough room for further actions)
Wireframe sketches - Notification
Smart Detection Interface Design
Actions that the user could do after detecting sensitive data:
(1) Replace with placeholder (2) Replace with fake data (3) Keep origin (4) Open settings
I explored how users' eyes would travel across different layouts
Wireframe sketches - Actions
Settings & Control
Built-in Rules
Pre-configured:
SSN, email addresses, phone numbers, IDs...etc.
Rule Cards:
Visual representation of each privacy rule
Custom Rule Creation
Users can define their own sensitive patterns with:
Rule naming
Regular expression patterns
Custom replacement text
Test field for validation
Wireframe sketches - Settings
Prototyping
Vibe-Coded interfaces to test the usability
Pattern B
The centered popup forces an immediate decision, completely breaking the user's writing flow. When users are in a creative mindset, this interruption can cause them to lose their train of thought entirely.
Pattern C
The corner-positioned notification maintains peripheral awareness without demanding immediate attention.
Pattern B feels like the system is "taking control" away from the user. Pattern C feels like the system is "offering helpful information" , a crucial psychological difference that reduces user frustration and increases long-term adoption.
Solution
Real-time Detection
Get instant alerts the moment you're about to share sensitive information, before you hit send.
Built-in Protection Rules
Comes preloaded with common privacy rules for US and global identifiers: SSNs, phone numbers, credit cards, and more.
Custom Rules for Any Data Type
Easily define your own detection patterns with regex, example data, and optional replacement text: no coding expertise required.
Quick Templates for Fast Setup
Start with ready-made templates for IDs, license plates, medical records, product serials, and other frequent formats.
Reflection
When AI Made Me Feel Unstoppable
Seeing real people download and use something I built feels like a dream. What started as me worrying about accidentally sharing work details with ChatGPT somehow turned into a little tool that might help others too.
With AI helping me code, I could try out ideas as actual working pieces. When something felt clunky, I could tweak it right away while I still remembered exactly what bugged me about it. Using my own creation taught me things I never anticipated. Living with the extension daily while building it showed me tiny moments of user frustration and delight that I probably would have missed otherwise. Like how grateful I felt when it quietly caught something risky, or how annoying it was when alerts popped up at the wrong moment.AI became this amazing creative partner. I could focus on the feelings and interactions I wanted to create while AI helped with the technical stuff I was still learning. It felt like having a really patient coding buddy who never made me feel silly for not knowing something.
The best part has been reading reviews from friends who find Privacy Guard genuinely helpful! It makes me want to keep exploring what else might be possible when you combine thoughtful design with AI assistance. There's something magical about turning "wouldn't it be nice if..." into something real that people actually use.
