CognitiveSense

Empower users against online manipulation. CognitiveSense detects deceptive tactics on shopping & social media using AI. 12 detectors, 4 APIs, 8 languages, real-time protection.

Comment

Inspiration

We noticed that online manipulation is everywhere—from fake countdown timers on shopping sites to emotional manipulation on social media. Most users don't realize how systematically they're being deceived. We wanted to build something that educates users in real-time, helping them recognize and resist these tactics. That's how CognitiveSense was born.

What it does

CognitiveSense is an AI-powered Chrome extension that detects and explains manipulation tactics on shopping and social media platforms in real-time.

Shopping Detection (6 tactics):

  • Urgency & Scarcity (countdown timers, "limited stock")
  • Anchoring & Price Manipulation (fake discounts)
  • Social Proof (fake reviews, suspicious ratings)
  • FOMO (exclusivity, artificial scarcity)
  • Bundling (forced bundles, hidden costs)
  • Dark Patterns (deceptive UI, hidden options)

Social Media Detection (6 tactics):

  • Misinformation (false/misleading information)
  • Emotional Manipulation (fear, anger, outrage triggers)
  • Echo Chambers (limited perspective diversity)
  • Fake Accounts (bot behavior, automated patterns)
  • Toxic Content (hate speech, harassment)
  • Political Manipulation (bias, propaganda)

Key Features:

  • Real-time detection with confidence scores
  • AI-generated explanations and user-friendly warnings
  • Multi-language support (8 languages)
  • Scroll-based re-analysis for social media feeds
  • Tab-aware panel (each tab maintains its own data)
  • Professional UI with floating badge and element overlays

How we built it

Architecture:

  • Agent-Based System: 2 specialized agents (Shopping & Social) that intelligently route to appropriate detectors
  • 12 Detectors: Modular detectors for each manipulation type, running in parallel
  • 4 Chrome AI APIs:
    • Prompt API for core detection logic
    • Writer API for user-friendly warnings and tips
    • Translator API for 8-language support
    • Summarizer API (initialized for future features)

Tech Stack:

  • TypeScript + React for UI
  • Chrome Extension APIs for browser integration
  • Per-URL caching for performance
  • Debounced scroll detection for social media

Key Implementation Details:

  • Content script analyzes page on load
  • PageContextBuilder extracts domain, URL, and content
  • AgentRegistry routes to appropriate agent based on page type
  • Detectors run in parallel for performance
  • Results stored per-URL for tab awareness
  • MultiLanguageManager handles language detection and translation

Challenges we ran into

  1. Agent Enablement Issue: SocialMediaAgent was initialized but disabled by default. Took debugging to realize the settings merge wasn't working for new agents. Fixed by implementing proper settings merging with defaults.

  2. Tab-Awareness Problem: Panel was showing data from wrong tabs because it was falling back to latestDetections. Solved by removing fallback logic and only showing URL-specific data.

  3. AIManager Access: Social detectors couldn't access the Prompt API because aiManager wasn't being passed to the context. Fixed by attaching aiManager to context before calling agent.detect().

  4. Mermaid Diagram Conflicts: Variable naming conflicts (SA vs SA_API) broke diagram rendering. Resolved by renaming to unique identifiers (ShopAgent, SocialAgent).

  5. Chrome AI API Availability: Language Detector API wasn't available, so we implemented script-based + keyword-based language detection as fallback.

Accomplishments that we're proud of

✅ 12 Fully Functional Detectors: All detectors working with real AI analysis, not just pattern matching

✅ 4 Chrome AI APIs Integrated: Most projects use 1-2 APIs; we integrated all 4 effectively

✅ Dual-Agent Architecture: Intelligent routing system that scales to new agents easily

✅ Multi-Language from Day 1: 8 languages with real-time translation, not just UI translation

✅ Real-Time Social Media Analysis: Scroll-based detection that re-analyzes new content as users browse

✅ Tab-Aware System: Each tab maintains independent data—no cross-tab pollution

✅ Professional UI: Floating badge, expandable panel, element overlays, language selector

✅ Demo Shopping Page: Beautiful showcase page demonstrating all 6 shopping detectors

✅ Production-Ready Code: Comprehensive error handling, logging, and graceful fallbacks

✅ Comprehensive Documentation: Video script, architecture diagrams, roadmap, submission form

What we learned

  1. Agent-Based Architecture Scales: Designing with agents from the start made adding SocialMediaAgent straightforward

  2. Settings Management is Critical: Merging defaults with stored settings is essential for backward compatibility

  3. Chrome AI APIs are Powerful: The combination of Prompt, Writer, and Translator APIs creates a compelling user experience

  4. Tab Awareness Matters: Users expect each tab to have independent state—this is non-negotiable

  5. Debouncing is Essential: For scroll-based detection, debouncing prevents excessive API calls

  6. Logging is Your Friend: Comprehensive console logging made debugging much faster

  7. User Experience > Raw Features: A well-designed UI with clear warnings is more valuable than detecting more tactics

What's next for CognitiveSense

Phase 2: Expand Detection Coverage

  • News Bias Agent (6 detectors for news sites)
  • Social Pulse Agent (6 detectors for emerging platforms)
  • Global expansion to 50+ sites

Phase 3: Advanced Features

  • Browser history analysis to identify manipulation patterns
  • User feedback system to improve detector accuracy
  • Community-driven detector library
  • Integration with fact-checking APIs

Phase 4: Enterprise & Education

  • B2B version for organizations
  • Educational dashboard for teaching media literacy
  • API for third-party developers
  • Chrome Web Store publication

Phase 5: AI Enhancement

  • Fine-tuned models for specific domains
  • Personalized detection based on user preferences
  • Predictive alerts for emerging manipulation tactics
  • Integration with more Chrome AI APIs as they become available

Long-term Vision: Make CognitiveSense the go-to tool for online safety, helping millions of users make informed decisions free from manipulation.

That's our story! CognitiveSense is ready to protect users from online manipulation. 🚀

Built With

Share this project:

Updates