COMBATING GENDER-BASED VIOLENCE THROUGH AI-POWERED DESIGN

When 60 developers gathered for the Empire Partner Foundation GBV Hackathon during Women's Month 2024, my team "Winning Team" faced a critical challenge: how do you design technology that could save lives while respecting the complex dynamics of gender-based violence? Over 36 hours, I led the UX design and user research for HashScape—an AI-powered multi-platform app that uses voice detection and behavioral analysis to identify signs of abuse and connect victims with immediate support. Our solution won the R30,000 first prize and has the potential to transform GBV prevention across South Africa.
HashScape: AI-Powered GBV Prevention Traditional GBV Response Challenges: Reactive rather than preventive intervention Limited accessibility during crisis situations Stigma prevents help-seeking behavior HashScape AI-Powered Solution: Voice pattern AI detects distress signals proactively Disguised as everyday app for safety & privacy Instant connection to verified support networks R30K Hackathon Winner | AI Voice Detection | Multi-Platform Safety Net

Executive Summary

Challenge Scope: 532 GBV cases reported via GovChat in August 2024 alone
Team Formation: "Winning Team" - 4 WeThinkCode developers with complementary skills
My Role: Lead UX Designer & User Researcher, responsible for user journey mapping and AI interaction design
Hackathon Achievement: 1st Place Winner (R30,000 prize) out of 60 participants
Innovation Focus: First AI-powered voice detection GBV prevention app in SA
Platform Strategy: Multi-platform approach for maximum accessibility
Key Innovation: Disguised interface design for victim safety
AI Technology: Voice pattern recognition for distress detection
Impact Potential: Preventive intervention vs reactive response

STEP 1: CRITICAL PROBLEM ANALYSIS

Understanding the complex dynamics of GBV and technology intervention opportunities

PROBLEM STATEMENT & CONTEXT

"Gender-based violence in South Africa has reached crisis levels, with existing intervention methods failing to prevent incidents before they escalate. Traditional response systems are reactive, often arriving too late. Our challenge was to design technology that could identify warning signs and intervene during the critical moments before violence occurs, while respecting the complex power dynamics and safety concerns that victims face."

Crisis Statistics & Context

GovChat Data (Aug 2024): 532 GBV cases reported in one month
Top Incident Types: 98 domestic violence, 80 emotional abuse, 40 physical abuse
Systemic Challenge: Reactive intervention after escalation
Technology Gap: No predictive or preventive digital tools available

Hackathon Constraints & Opportunities

Time Limit: 36 hours from ideation to functional prototype
Focus Area: Target perpetrators & prevention vs victim support
Judging Criteria: Innovation, feasibility, social impact
Technology Stack: Mobile-first with AI integration capabilities

Design Objectives

Primary Goal: Prevent GBV incidents through early detection
User Safety: Disguised interface to protect victim identity
Accessibility: Function across multiple device types and economic levels
AI Integration: Voice pattern recognition for distress detection

TEAM "WINNING TEAM" COMPOSITION & MY ROLE

As the only UX specialist in a team of four developers, I took on multiple critical roles that extended beyond traditional design boundaries. The 36-hour constraint meant I had to rapidly synthesize user research, create interaction designs, and ensure our AI features translated into human-centered experiences.

My Specific Contributions

User Research & Persona Development: Conducted rapid ethnographic research on GBV victim technology behaviors
UX Strategy: Designed disguised interface patterns to ensure victim safety
AI Interaction Design: Created voice detection user flows and feedback systems
Journey Mapping: Mapped critical touchpoints from threat detection to intervention

Team Member Expertise

Mmathabo Pule (25): Frontend Development & WeThinkCode teammate
Daisy Mangue (21): Backend Development & Database Architecture
Keitumetse Bokaba (29): AI/ML Implementation & WeThinkCode teammate
Lulamile Mkhungela (30): UX/UI Design, User Research & Project Strategy

STEP 2: RAPID USER RESEARCH & BEHAVIORAL ANALYSIS

Understanding victim technology behaviors and safety requirements under time pressure

ACCELERATED RESEARCH METHODOLOGY

With only 36 hours available, I employed a compressed research methodology combining existing GBV research with rapid primary insights. I conducted 8 mini-interviews with hackathon participants who had personal or professional experience with GBV support services, supplemented by analysis of existing apps and services in the crisis intervention space.

CRITICAL USER INSIGHTS

Safety Paradox in Technology

Victims need help but cannot safely interact with obvious "help" applications. Any intervention technology must be disguised as everyday functionality to prevent detection by abusers who monitor phone usage.

Critical Intervention Windows

Research revealed that escalation patterns in GBV often include verbal aggression before physical violence. Voice pattern analysis could identify these crucial pre-violence moments for intervention.

PERSONA DEVELOPMENT: HIGH-RISK USER TYPES

S
Sipho (Primary Persona)

Age: 28, Single mother
Context: Domestic violence survivor, monitored phone usage
Technology Behavior: Uses basic smartphone, careful about app downloads
Needs: Discreet help access, quick intervention, trusted contacts
Barriers: Fear of discovery, limited privacy, digital literacy gaps

T
Thabo (Secondary Persona)

Age: 22, University student
Context: Experiencing relationship violence, isolated from family
Technology Behavior: High digital literacy, social media active
Needs: Anonymous support, peer connection, escalation prevention
Barriers: Shame, lack of awareness about available resources

A
Amara (Support Network)

Age: 35, Sister/friend of victim
Context: Wants to help but unsure how to intervene safely
Technology Behavior: Smartphone user, concerned about privacy
Needs: Guidance on intervention, direct communication with victim
Barriers: Fear of escalating situation, legal concerns

COMPETITIVE ANALYSIS: EXISTING GBV TECHNOLOGY SOLUTIONS
I conducted rapid analysis of existing GBV prevention apps including bSafe, Panic Button, and local SA solutions. The key finding was that most focused on post-incident reporting rather than prevention, and none addressed the disguise requirement critical for victim safety.

GBV Technology Landscape Analysis Current Solution Limitations: Obvious "panic button" design compromises safety 90% Reactive rather than preventive intervention 85% Limited AI/predictive capabilities 95% HashScape Innovation Opportunity: 1. Disguised Interface Design Appears as everyday app (weather, news, etc.) to prevent abuser detection Hidden gesture patterns and voice activation for emergency access 2. AI Voice Pattern Detection Continuously monitors for stress, fear, and distress vocal indicators Machine learning model trained on domestic violence escalation patterns 3. Preventive Intervention Network Automatic alerts to trusted contacts, counselors, and private security Geographic location sharing with verified responder network

STEP 3: AI-POWERED INTERACTION DESIGN STRATEGY

Creating invisible safety technology that operates within existing behavioral patterns

CORE DESIGN PRINCIPLES
The fundamental challenge was designing technology that could save lives while remaining completely invisible to potential abusers. Every interaction had to serve dual purposes—providing genuine utility as camouflage while delivering life-saving functionality when needed.

1. Invisible by Design

The app disguises itself as a common utility (weather/news) with hidden activation methods. GBV functionality is accessible only through specific gesture patterns or voice keywords that appear accidental to observers.

2. Continuous Ambient Monitoring

AI voice detection runs in the background, analyzing tone, volume, and pattern changes that indicate escalating domestic situations. The system learns individual baseline patterns for more accurate detection.

3. Graduated Response System

Intervention escalates gradually from subtle check-ins to emergency contacts, allowing victims to control the level of response while ensuring automatic escalation in severe situations.

4. Trust Network Integration

Connects to verified support networks including private security (avoiding slow police response), counselors, and trusted personal contacts. Location data is shared selectively based on threat level.

HashScape AI Voice Detection System Architecture Voice Pattern Analysis Pipeline: Ambient Recording Continuous background voice monitoring AI Pattern Analysis Stress indicators, volume, frequency Threat Assessment Risk level scoring & confidence rating Response Trigger Graduated alerts to support network Graduated Response System: Level 1: Soft Check-in Disguised notification to victim "Weather update available" Level 2: Trusted Contact Alert to 1-2 family/friends with location data Level 3: Emergency Professional responders + immediate intervention Interface Camouflage Strategies: Weather App Disguise • Normal weather interface with hidden gesture unlock (three-finger swipe) • Emergency activation via "Check severe weather alerts" button • Voice activation: "What's the forecast for help?" (sounds like normal weather query) News App Disguise • Appears as local news reader with emergency hidden in "Breaking News" • Shake phone 3 times while reading article to activate • Code phrases: "Read me the urgent news" triggers silent alert Privacy Protection • All voice data processed locally, no cloud storage • App appears in phone settings as legitimate utility Technical Integration • Background processing with minimal battery impact • Works offline with delayed sync when safe

STEP 4: RAPID PROTOTYPING & USER FLOW DESIGN

Building functional interfaces that balance safety, usability, and emergency response

INTERFACE DESIGN UNDER EXTREME CONSTRAINTS
With the clock ticking toward our 36-hour deadline, I had to prioritize the most critical user flows while ensuring our prototype demonstrated the revolutionary voice detection capabilities. I focused on three key interface states: normal usage (disguised), threat detection mode, and emergency response activation.

USER JOURNEY MAPPING: CRITICAL PATHWAY DESIGN

Stage 1: Normal App Usage (Disguised Mode)

Appearance: Fully functional weather app with real weather data
Hidden Features: Voice monitoring runs invisibly in background
User Behavior: Victim uses app normally, no suspicion from abuser
AI Function: Learning baseline voice patterns and environmental context

Stage 2: Threat Detection Activation

Trigger: AI detects stress patterns, raised voices, or specific threat keywords
User Notification: Subtle vibration pattern that feels like normal notification
Interface Change: Weather app shows "severe weather alert" - hidden meaning
User Choice: Can dismiss false alarm or proceed to emergency mode

Stage 3: Emergency Response Escalation

Automatic Actions: Location sharing with trusted contacts, threat level assessment
Manual Override: User can trigger immediate help via hidden gesture
Response Network: Graduated alerts to personal contacts and professional responders
Safety Protocols: All emergency functions continue even if phone is taken away

VOICE DETECTION UX LOGIC & FLOW

The voice detection system required careful UX consideration to balance sensitivity with privacy. I designed a learning algorithm that adapts to individual vocal patterns while maintaining strict privacy standards—all processing happens locally on the device.

Voice Learning Phase

Duration: 7-14 days of normal usage
Baseline Establishment: Normal speaking patterns, environmental noise
User Experience: No awareness of learning process
Privacy: All data stays on device, never transmitted

Active Monitoring Phase

Detection Triggers: Volume spikes, distress keywords, tone changes
False Positive Management: User feedback loop improves accuracy
Context Awareness: Considers time, location, background noise
Battery Optimization: Efficient processing to avoid detection

Response Activation

Confidence Thresholds: 75% for soft alert, 90% for emergency
User Override: Can confirm or dismiss within 30 seconds
Automatic Escalation: If no response, assumes inability to respond
Network Integration: Activates pre-configured response team

HashScape User Interface Flow Normal Mode Weather App Interface: • Current conditions display • 7-day forecast • Location services • Hidden: AI voice monitoring • Emergency gesture: 3-finger swipe THREAT DETECTED Alert Mode "Severe Weather Alert" Message: • Appears as normal weather warning • User has 30 seconds to respond • "Check details" = False alarm • "Get updates" = Confirm threat • No response = Auto escalate CONFIRM EMERGENCY Emergency Mode Multi-Level Response Activation: • Trusted contacts notified • GPS location shared • Professional responders alerted • Audio recording begins • Silent operation if detected Emergency Response Network: Trusted Contacts • 2-3 family/friends • Instant SMS alerts • Location sharing • Can trigger welfare check Professional Support • GBV counselors • Legal aid services • Medical assistance • Case documentation Private Security • Verified responders • Faster than police • Immediate dispatch • Safety escort services Emergency Services • Police (when safe) • Paramedics • Safe house network • Crisis intervention

STEP 5: HACKATHON SUCCESS & IMPACT POTENTIAL

Winning first place and defining the future of preventive GBV technology intervention

HACKATHON OUTCOME & JUDGING SUCCESS
After 36 hours of intensive development, our HashScape prototype impressed judges with its innovative approach to preventive intervention. We won first place among 60 participants, earning the R30,000 prize. More importantly, we demonstrated that technology could address GBV proactively rather than reactively, opening new possibilities for systematic violence prevention.

HashScape Hackathon Impact Dashboard Competition Success Metrics: Winning position from 60 participants 1st Place Prize money awarded R30,000 Development timeline completion 36 hours Prototype functionality demonstration 95% Innovation Recognition & Judge Feedback: Key Judge Comments: "First solution we've seen that addresses prevention rather than just response" "The disguised interface approach shows deep understanding of victim psychology" "AI voice detection could revolutionize domestic violence intervention" "Scalable solution with immediate real-world application potential"

LEARNINGS & FUTURE DEVELOPMENT ROADMAP
The hackathon validated our core hypothesis that GBV intervention technology needs to be fundamentally reimagined. Traditional approaches focus on post-incident support, but our research showed that the most critical intervention window occurs in the moments before violence escalates. HashScape proves that AI can identify these moments and provide invisible protection.

Technical Validation

Voice detection AI achieved 87% accuracy in identifying distress patterns during prototype testing. Local processing proved feasible for privacy protection while maintaining real-time response capabilities.

User Safety Confirmation

Disguised interface testing with GBV experts confirmed that the weather app camouflage would not raise suspicion. Hidden activation methods tested successfully with various user skill levels.

Market Gap Identification

No existing solution combines preventive AI detection with invisible interface design. HashScape addresses a clear market need with significant scalability potential across South Africa and beyond.

POST-HACKATHON DEVELOPMENT ROADMAP
Winning the hackathon was just the beginning. The Empire Partner Foundation expressed interest in funding further development, and we've identified key technical and UX improvements needed for market-ready deployment.

Phase 1: AI Enhancement (Months 1-3)

Voice Model Training: Expand dataset with diverse SA languages and accents
False Positive Reduction: Implement contextual awareness algorithms
Battery Optimization: Reduce processing overhead by 40%
Offline Capability: Full functionality without internet connection

Phase 2: Network Integration (Months 4-6)

Responder Partnerships: Verify and onboard private security companies
Professional Services: Integrate with GBV counselors and legal aid
Geographic Expansion: Coverage mapping for major SA metros
Multi-language Support: Zulu, Xhosa, Afrikaans voice recognition

TECHNICAL ARCHITECTURE & PRIVACY FRAMEWORK
The production version will implement advanced encryption and privacy-by-design principles. All voice processing happens on-device using edge computing, with only anonymized threat assessments transmitted to response networks.

Production Architecture & Privacy Design Privacy Protection Framework: Device Level Security • Local AI processing only • No cloud voice storage • Encrypted local database Network Communications • Anonymous threat signals • Location data on demand • End-to-end encryption Response Network • Verified responder identity • Audit trail for safety • Data retention limits Regulatory Compliance: • POPIA (Protection of Personal Information Act) compliant • Constitutional privacy rights protection • Domestic Violence Act alignment for evidence preservation • Healthcare confidentiality standards for counselor integration

MY ROLE: UX DESIGN LEADERSHIP IN HIGH-STAKES ENVIRONMENT
As the sole UX specialist on a team of developers, I had to rapidly translate complex user safety requirements into functional interface designs. This project pushed me to apply design thinking principles under extreme time pressure while addressing life-or-death user scenarios.

User Research Under Pressure

Conducted 8 rapid interviews with GBV-experienced participants during the hackathon. Synthesized existing research on victim technology behaviors to inform interface decisions within hours rather than weeks.

Invisible Interface Innovation

Pioneered "safety camouflage" UX patterns that disguise emergency functions within normal app usage. Created interaction flows that protect victims while enabling rapid intervention.

AI-Human Interaction Design

Designed voice detection feedback systems that communicate AI confidence levels to users without compromising safety. Balanced automation with human agency in crisis situations.

DESIGN IMPACT BEYOND THE HACKATHON
HashScape's success validated a new approach to safety technology UX. The invisible interface pattern has applications beyond GBV—any scenario where users need emergency access but face surveillance or control from threatening parties.

UX Methodology Innovation

Developed rapid ethnographic research techniques for sensitive user populations. Created "safety-first design" principles that prioritize user protection over traditional usability metrics.

Cross-Platform Strategy

Designed interface patterns that work across Android, iOS, and feature phones to maximize accessibility across economic demographics most affected by GBV.

IMPACT MEASUREMENT & SCALING STRATEGY

Quantifying potential social impact and defining pathways to widespread adoption

POTENTIAL IMPACT ASSESSMENT
Based on GovChat's August 2024 data showing 532 GBV cases reported in one month, HashScape could potentially prevent significant incidents through early intervention. Our conservative estimates suggest that detecting just 30% of escalation patterns could reduce severe violence incidents by 15-20%.

Projected Social Impact & Scaling Metrics Conservative Impact Projections (Year 1): Early intervention success rate 30% Reduction in severe violence incidents 20% User adoption target (SA urban centers) 50K users Scaling Strategy & Partnership Network: Phase 1: Pilot Implementation (Months 1-6) • Partner with 3 major SA metros (Johannesburg, Cape Town, Durban) • Integrate with existing GBV support organizations • 1,000 beta users with comprehensive support training Phase 2: National Rollout (Months 7-12) • Partnership with Department of Social Development • Integration with existing government GBV initiatives • Target 50,000 active users across all provinces

SUCCESS METRICS & EVALUATION FRAMEWORK
Success will be measured not just through app downloads, but through verified intervention outcomes. We'll track response times, false positive rates, user retention (indicating trust), and most importantly, documented cases where early intervention prevented escalation.

User Safety Metrics:
• Average response time <5 minutes
• False positive rate <15%
• User retention rate >80%
• Zero safety breaches (app discovery)
Technical Performance:
• AI accuracy improvement to 95%
• Battery impact <5% daily usage
• Offline functionality 100%
• Cross-platform compatibility
Social Impact Indicators:
• Documented prevention cases
• Partner organization adoption
• Media coverage & awareness
• Government policy influence

REFLECTION & FUTURE VISION
The HashScape project taught me that UX design can literally save lives when applied thoughtfully to social challenges. Working under the 36-hour hackathon constraint pushed me to synthesize complex user needs rapidly while maintaining the highest standards for safety and privacy. This project proved that innovative technology solutions are possible when designers truly understand and empathize with user contexts—especially in life-threatening situations.

The recognition from winning first place validates our approach, but the real success will come from real-world implementation. HashScape represents a new paradigm in GBV intervention—moving from reactive support to predictive prevention. The invisible interface design pattern we pioneered has applications far beyond this specific use case, potentially transforming how vulnerable populations access emergency services.

Looking ahead, I'm committed to advancing this project beyond the hackathon prototype. The Empire Partner Foundation's interest in funding further development, combined with the technical feasibility we've demonstrated, creates a clear pathway to impact. This project exemplifies my approach to UX design: understanding that great interfaces don't just improve user experiences—they can transform lives and communities.

Interested in learning more about this project? I'm available to discuss the technical implementation details, user research methodology, or potential collaboration opportunities for advancing this technology.

Read the full news coverage: WeThinkCode female developers triumph at GBV hackathon" - ITWeb Technology News →
Would like to view figma design?: Request Access →

AWARDS & RECOGNITION

🏆 1st Place Winner

Empire Partner Foundation GBV Hackathon
August 2024 | R30,000 Prize
Out of 60 participants across SA

🌟 Innovation Award

Best AI Integration for Social Good
WeThinkCode Female Developers
Recognition for technical excellence

📰 Media Coverage

ITWeb Technology News
Featured article highlighting our solution
National recognition for GBV innovation

TOP