Haven

Hidden in plain sight. Heard when it matters most.

Comment

1. Project Title and Description

Project Title: Haven– Hidden in plain sight. Heard when it matters most.

Description:
Haven is a secure, privacy-first platform that empowers survivors of domestic violence and gender-based abuse to seek help discreetly. Using AI-powered steganography, victims can hide urgent SOS messages inside ordinary photos and post them on any social media without raising suspicion. An empathetic 3D AI avatar provides 24/7 mental health support in multiple languages, while a smart legal assistant delivers instant, jurisdiction-specific guidance. Built for students by students, Haven turns silent suffering into safe, actionable strength connecting victims to help without alerting abusers.

2. Problem Statement

Gender-based violence remains one of the most pervasive human rights crises worldwide. According to the latest UNODC–UN Women data (2025), approximately 50,000 women and girls were killed by intimate partners or family members in 2024 alone an average of 137 women and girls every single day. The World Health Organization reports that nearly 1 in 3 women (840 million globally) have experienced physical or sexual intimate-partner violence in their lifetime.

In India, the National Family Health Survey (NFHS) shows that nearly one in three ever-married women (32%) has faced physical, sexual, or emotional spousal violence. Official NCRB records show over 445,000 crimes against women reported annually, yet experts agree the true figure is far higher due to fear, stigma, economic dependence, and surveillance by abusers who monitor phones and social media. Victims often cannot make calls, send obvious messages, or openly seek legal or mental health help without risking escalation.

Existing safety apps rely on visible buttons or loud alerts which are features that abusers can easily detect and disable. There is a critical gap for truly discreet, multi-layered support that protects privacy while delivering immediate emotional, legal, and safety resources.

3. Solution Overview

Haven solves this through three seamlessly integrated, privacy-first features:

  • Discreet SOS via AI Steganography
    Users type a short trigger phrase → AI expands it into a full, context-aware distress message → the message is invisibly encoded inside any normal-looking photo. The photo can be posted on Instagram, WhatsApp, or Facebook. Trusted contacts or authorities scan the image with Haven’s decoder to retrieve the exact location, message, and emergency details completely invisible to anyone else.

  • Empathetic AI Mental Health Companion
    A lifelike 3D talking avatar (with lip-sync and emotional expressions) offers 24/7 private conversations, coping strategies, safety planning, and de-escalation techniques. Supports English, Hindi, and Tamil (expandable to more languages).

  • Intelligent Legal & Resource Assistant
    Powered by vector search over official legal documents, the system instantly provides jurisdiction-specific guidance (Indian laws + global resources) and connects users to verified helplines, shelters, and NGOs.

The entire system uses end-to-end encryption and includes a “panic mode” that erases traces if needed.

4. Technical Details

Architecture: Modern full-stack with separate frontend and backend for scalability and security.

  • Frontend: Next.js 15 (App Router) + TypeScript + Tailwind CSS + shadcn/ui + Three.js / Ready Player Me for 3D avatar
  • Backend: FastAPI (Python) with JWT authentication
  • AI/ML Layer:
    • Gemini 1.5 Flash / Groq (Llama-3) for message expansion & legal chat
    • Sentence Transformers + MongoDB Atlas Vector Search for legal document retrieval
    • ElevenLabs or OpenAI TTS for avatar voice & lip-sync
  • Steganography: Custom Python implementation using Pillow + LSB (Least Significant Bit) encoding
  • Database: MongoDB Atlas (free tier) with Vector Search
  • Deployment: Vercel (frontend) + Render / Railway (backend) + AWS S3 for temporary image handling
  • Security: End-to-end encryption, rate limiting, and optional self-destruct mode

The stack is student-accessible yet production-grade, focusing on clarity, speed, and privacy.

5. Use Case and Impact

Primary Use Case:
A woman experiencing domestic abuse opens Haven on her phone, types “help me” while pretending to browse photos. The app encodes her precise location + message into a beautiful landscape photo, which she posts normally. Her trusted contact or local women’s helpline instantly decodes it and responds without the abuser knowing anything happened.

Additional Use Cases:

  • College students in hostels using the AI companion for emotional support.
  • NGOs using the dashboard to track anonymized severity trends.
  • Global travelers accessing region-specific legal help.

Expected Impact:

  • Reduce response time from hours/days to minutes by removing the “visible alert” barrier.
  • Lower psychological burden through always-available, non-judgmental AI support.
  • Increase reporting and help-seeking rates (currently only ~10–14% of victims seek formal help).
  • Scalable to any country by simply adding more legal documents to the vector database.

Long-term, Haven can integrate with national helplines (e.g., India’s 181) and generate anonymized data to help governments and NGOs design better prevention programs.

Built With

Share this project:

Updates