🧠 Bifocal – AI for Early Detection of Emotional & Psychological Abuse

💡 Inspiration

Domestic abuse and harassment often hide behind subtle messages—gaslighting, manipulation, and psychological control. Traditional tools fail to detect these patterns early, leaving individuals vulnerable and unsupported. Bifocal was created to fill this gap: a dual-lens AI platform that analyzes conversations from both the victim and offender perspectives, highlighting escalating risks through DSM-informed insights and empowering users with actionable support.

🚀 What It Does

Bifocal uses advanced AI and behavioral analysis to assess message history and flag potential risks, including:

• 🧠 Victim Lens: Detects trauma language, emotional distress, and psychological harm
• ⚠️ Offender Lens: Flags signs of coercive control, narcissism, and antisocial tendencies
• 💬 Sentiment & Pattern Analysis: Uses DSM-5-informed NLP to detect red flags and emotional shifts
• 🧍 Personalized Guidance: Offers resources, explanations, and steps tailored to the user’s needs
• 🗣️ User Feedback Integration: Allows users to highlight harmful messages for deeper analysis

🛠️ How We Built It

• Frontend: React + TypeScript + Emotion for a calming, responsive UI

• NLP & ML:

  • Sentiment analysis
  • Entity recognition
  • Linguistic + psychological pattern detection
  • Risk scoring and escalation modeling
  • Anomaly detection for behavior shifts

• Frameworks:

  • Retrieval-Augmented Generation (RAG) with DSM-5 context modeling
  • MCP (Model Context Protocol) to securely access iMessage data
  • Clinical criteria mapped to psychological signals for high-fidelity analysis

🧩 Challenges We Faced

• Context Understanding: Handling nuance in tone, slang, and emotional ambiguity across varied messages
• Secure Data Access: Establishing private, ethical access to iMessage logs with strong privacy protocols

✅ Accomplishments

• Developed a dual-lens model that evaluates both victim and offender behaviors
• Created a clean, approachable UI for communicating complex psychological insights
• Achieved high pattern detection accuracy validated against DSM-5 criteria

📚 What We Learned

This project deepened our knowledge of:

• Behavioral psychology and emotional abuse detection
• NLP use in high-sensitivity domains
• Frontend empathy-driven design for users navigating trauma
• Secure, ethical handling of sensitive conversation data

🔮 What’s Next for Bifocal

• 📱 Launching a mobile app for broader reach
• 🌐 Integrating platforms like WhatsApp and social media
• 🧠 Enhancing message context recognition (friend vs. stranger tone)
• 🕒 Adding daily emotional check-ins to monitor user well-being

🖼️ Screenshots

🛠️ Built With

• AI & ML: huggingface, bert, RAG, machine-learning, python
• Frontend: react, typescript, css, html, emotion, javascript
• Speech & NLP: google-web-speech-api, natural-language-processing
• Infrastructure: sql, mcp

---

📬 Contact

• GitHub: @SecretariatV
• Email: oliver.b25.f@gmail.com
• Telegram: @ares_orb
• Twitter (X): @OVB_Coder