SafeGuard AI

SafeGuard AI — Empowering safer digital spaces with trust, technology, truth, and accountability.

Comment

SafeGuard AI

Intelligent Real-Time Harassment Detection & Evidence System
Building Trust for the Digital Economy

Overview

SafeGuard AI is an intelligent real-time harassment detection system that protects users from online threats by combining Artificial Intelligence and Blockchain technology. It detects, records, and prevents online harassment while providing victims with tamper-proof evidence, fostering safer digital environments.

Quick Stats

Metric Value
Detection Speed < 500ms per message
Accuracy 93%+ on threat detection
Blockchain 100% tamper-proof evidence
Languages English

Average Detection Speed: ~500ms per message; longer messages may take slightly longer depending on length.

Problem Statement

The Crisis

Online harassment and cyberbullying are rising in India, particularly affecting women and marginalized groups. Many incidents go unreported due to the lack of reliable evidence collection, leading to:

  • Reduced digital participation and engagement on social media, e-commerce, and other online platforms.
  • Productivity losses for content creators, freelancers, and online entrepreneurs.
  • Increased moderation and legal costs for platforms.

Economic Consequences

Unsafe online environments have real economic impacts:

  • User churn & lost revenue: Harassment drives users to reduce activity or leave platforms, directly lowering engagement, ad revenue, and transaction volume.
  • Platform costs: Manual moderation, legal compliance, and evidence handling are expensive.
  • Reduced productivity: Online creators and digital workers lose income and engagement opportunities.

SafeGuard AI addresses these issues by enabling safer digital environments, improving user trust, protecting economic activity, and reducing platform costs.

Solution

SafeGuard AI provides a three-layer protection system:

  1. AI Detection Layer

    • Pre-trained Toxic-BERT model for real-time analysis (< 500ms)
    • Detects sexual harassment, violent threats, hate speech, abusive language
    • Categorizes severity: HIGH, MEDIUM, LOW

  2. Blockchain Evidence Layer

    • Logs every threat to an immutable blockchain
    • Generates tamper-proof evidence for legal use
    • Stores timestamp, threat type, severity, and content hash
    • Victims can download court-ready reports

  3. Pattern Detection Layer

    • Identifies coordinated harassment campaigns
    • Links multiple accounts targeting the same user
    • Detects escalation patterns and sends alerts for organized attacks

Key Features

Feature Highlights
Beautiful UI/UX Modern gradient design, responsive layout, intuitive navigation, dual views (Post Owner & Commenter)
AI-Powered Detection Real-time threat analysis, multi-category classification, confidence scoring
Blockchain Security Immutable logging, SHA-256 hashing, chain verification, built-in Blockchain Explorer
Smart Alerts Severity gauges, real-time notifications, coordinated attack warnings
Analytics Dashboard Live threat stats, type distribution, timeline visualizations, Plotly graphs, system metrics
Evidence Export Download threat reports (CSV), court-ready documentation, blockchain proof included

Tech Stack

Layer Technologies & Notes
Backend Python, Pandas, Hashlib, Transformers (Hugging Face), PyTorch
Frontend Streamlit, Plotly, Custom CSS
AI/ML Pre-trained Toxic-BERT, NLP techniques, real-time threat detection
Blockchain Custom Python implementation, SHA-256 hashing, chain integrity verification

Economic & Social Impact

SafeGuard AI directly improves the digital economy by:

  • Boosting Platform Participation: Safer environments encourage more users, especially women, to stay active on social media, e-commerce, and digital services.
  • Reducing Moderation Costs: Automates detection and evidence logging, potentially saving millions in legal and human moderation expenses.
  • Protecting Productivity: Minimizes stress and work disruption for online creators and freelancers.
  • Enabling Legal & Policy Action: Provides court-ready evidence that strengthens compliance and accountability.

Quantitative Estimates (Example Impact)

These are example projections to illustrate potential economic benefits:

  • Moderation Cost Reduction: Automated harassment detection could potentially reduce moderation costs by 30–40% per 1M active users by reducing some manual processes, allowing human moderators to focus on complex cases.

  • Increased Engagement & Transactions: Safer online environments may lead to 5–10% higher transaction volume in e-commerce or digital marketplaces, as users feel more confident interacting and transacting.

  • Reduced User Churn: By minimizing harassment, platforms can help more creators and professionals remain economically active online, sustaining engagement and productivity.

Note: These figures are illustrative projections based on logical assumptions about user behavior and platform operations. Actual results may vary depending on implementation and platform size.

Personal Contributions and Learning

  • Learned about BERT and real-time harassment detection.
  • Studied coordinated attack detection linking multiple accounts.
  • Created interactive graphs using Plotly for attack patterns and threat statistics.
  • Assisted with UX/UI design using Claude AI.
  • User testing: Preeti Va acted as a user tester, providing feedback on usability, interface clarity, and overall experience, helping refine SafeGuard AI’s design.

Future Scope

  • Expand detection to Indian regional languages (Hindi, Tamil, Bengali, Marathi, Telugu)
  • Integrate with helplines, NGOs, and law enforcement for direct reporting
  • Train custom datasets to improve subtle harassment detection
  • Include dashboards showing economic cost savings and platform efficiency metrics

References

  • UN Women: Online Harassment and Digital Safety Reports
  • Pew Research: Online Harassment and Participation Studies
  • Academic papers on cyberbullying, NLP detection models, and blockchain-based evidence systems
Share this project:

Updates