SheShield AI

Inspiration

Every day, millions of women across the world experience harassment, unsafe commutes, and threatening situations — on public transport, walking home at night, on college campuses, and in workplaces.

What moved us most was not just the danger itself, but the helplessness that comes with it.

Most women already carry safety apps. But when a real moment of fear arrives — heart racing, someone following you, a threatening situation unfolding — those apps fail at the worst possible moment. You have to unlock your phone, find the app, navigate to a button, and press it.

That delay can cost everything.

We were also moved by something quieter — the fact that many women never report incidents or warn others because they fear judgment, disbelief, or social consequences. Dangerous areas stay dangerous because warnings never spread fast enough.

We asked ourselves one question:

What if technology could act faster than fear? What if a community of women could protect each other in real time?

That question became SheShield AI.

What it does

SheShield AI is a community-driven, AI-powered safety ecosystem that protects, supports, and empowers women — everywhere in the world.

🆘 Triple Tap SOS

Tap your screen 3 times — even from a locked phone — and instantly:

Your live GPS location is shared with all your emergency contacts
A real email alert is sent with a live tracking link
Your Safety Map activates in emergency mode

💬 Safe Chat — AI Distress Detection

An AI companion powered by Google Gemini + LangChain that is always listening. If you type "I feel unsafe", the system automatically:

Detects the distress signal without any button press
Alerts your emergency contacts immediately
Identifies nearby Shield community members
Opens your Safety Map with safe routes

🗺️ Real-Time Safety Map

An interactive map (Leaflet + OpenStreetMap) showing your live location, safe zones, danger hotspots, and nearby Shield members — working in any city, any country.

👥 Community Safety Network

Women share real-time safety alerts, incident reports, and location warnings with each other. These posts are read by our AI and converted into structured threat intelligence automatically.

🚨 AI Hotspot Danger Wire

An autonomous LangChain agent reads every community post and generates a live Emergency Briefing — extracting location, hazard type, and safety recommendations — 24 hours a day, 7 days a week.

🌐 Live Web Safety Intelligence

Search any city in the world — London, Lagos, São Paulo, Jakarta — and our AI fetches live safety data, recent incidents, and government advisories using Tavily Search.

📊 Wellness & Mood Tracker

An AI wellness tracker that monitors mood, energy, and cycle patterns — giving personalized suggestions because safety is not just physical, it is emotional too.

How we built it

SheShield AI is a full-stack, production-grade application built end-to-end during this hackathon.

Frontend

Built with Next.js 16, React 19, Tailwind CSS v4, and Framer Motion for smooth, trustworthy interactions designed specifically for high-stress moments.

AI Architecture — Decentralized LangChain Cluster

We built three autonomous AI agents running in parallel:

┌─────────────────────────────────────────┐
│         SheShield AI Hub                │
│    Decentralized LangChain Cluster      │
└──────────────┬──────────────────────────┘
               │
   ┌───────────┼───────────┐
   │           │           │
Safe Chat   Hotspot     Web Search
  Agent     Analyzer      Agent
(Gemini)  (LangChain)  (Tavily)

Agent	Role
Safe Chat Agent	Detects distress in natural language, triggers emergency protocol
Hotspot Analyzer	Reads community posts, extracts structured threat intelligence
Web Search Agent	Fetches real-time city safety data from the web

Real-Time Layer

A custom Socket.IO server handles live location broadcasting, SOS alerts, and community messaging in real time.

Tech Stack Summary

Layer	Technology
Frontend	Next.js 16 + React 19
Styling	Tailwind CSS v4 + Framer Motion
AI & Agents	LangChain + Google Gemini
Web Search	Tavily Search API
Real-time	Socket.IO
Database	MongoDB + Prisma ORM
Maps	Leaflet.js + OpenStreetMap
Email Alerts	Nodemailer
Auth	bcryptjs
Validation	React Hook Form + Zod
Language	TypeScript (strict)

Challenges we ran into

1. AI Distress Detection Accuracy

Getting the AI to reliably detect distress without over-triggering was the hardest problem we faced. Too sensitive and users get false alerts constantly. Not sensitive enough and the system misses real emergencies.

We iterated through dozens of prompt variations and test cases — casual mentions of tiredness vs genuine fear, indirect language, code-switching between languages — before reaching a balance we were confident in.

2. Real-Time Location Under Browser Constraints

Browsers require user permission for geolocation, and many users block it. We built a multi-layer fallback system:

If GPS is available → precise location shared instantly
If GPS is blocked → alert still fires, contacts are notified, user is asked to share location manually

The SOS alert always fires. Even without GPS.

3. Community Data → AI Intelligence Pipeline

Building the real-time pipeline from a raw community post → LangChain agent analysis → structured hotspot briefing was architecturally complex. We had to balance speed, accuracy, and cost — processing community posts efficiently without introducing dangerous delays.

4. Designing for Fear

Most products are designed for calm, focused users. SheShield is designed for someone who is scared.

Every decision — button size, message tone, map colors, alert wording — had to be tested against one question: "Would this work if I was terrified right now?"

This completely changed how we think about UX and product design.

5. Making It Truly Global

Early versions were unconsciously region-specific in language, examples, and data framing. We deliberately rebuilt the AI prompts, map sources, and search context to work equally well in every country — no assumptions about geography, language, or infrastructure.

Accomplishments that we're proud of

✅ Real SOS emails delivered — not a prototype, not a mockup. Emergency contacts receive actual emails with live tracking links within seconds of SOS trigger. We tested it. It works.
✅ AI distress detection with no button press — a user simply typing "I feel unsafe" triggers the full emergency protocol automatically. This is the core breakthrough of our system.
✅ Three autonomous AI agents working in parallel — Safe Chat, Hotspot Analyzer, and Web Intelligence all run simultaneously on our LangChain cluster, each serving a different safety function.
✅ Global coverage from day one — OpenStreetMap, Tavily web search, and language-agnostic AI prompts mean SheShield works in any city, any country, any language — with no additional configuration.
✅ Full-stack production build in a hackathon window — real-time Socket.IO, MongoDB, LangChain agents, live maps, email alerts, community feed, wellness tracker — built, integrated, and working end-to-end.

What we learned

Technically:

Orchestrating multiple LangChain agents in parallel, sharing context without conflicts, is a non-trivial distributed systems problem — and we solved it
Real-time geolocation systems must be designed for failure first — graceful degradation is not optional, it is the core requirement
Safety-critical AI prompts require a completely different standard of testing than typical LLM applications

As builders:

The best safety tool is one that disappears — it should not demand your attention during a crisis, it should just act
Community is infrastructure — the social layer of SheShield is as technically important as any AI feature
Designing for high-stress, high-fear moments is one of the most demanding UX challenges that exists — and one of the most meaningful

The biggest lesson:

When you build for the most vulnerable moment in someone's life, every second, every word, and every tap matters more than in any other product you will ever make.

What's next for SheShield AI

📱 React Native mobile app — iOS & Android, with background SOS and push notifications
⌚ Smartwatch integration — tap-to-SOS directly from Apple Watch or Wear OS
🗺️ City-level danger heatmaps — historical hotspot data visualized across entire cities
🔊 Voice-activated SOS — "SheShield, help me" triggers the full emergency protocol hands-free
🌐 Multi-language support — AI companions and community feeds in local languages
🏫 Campus & corporate partnerships — deploying SheShield as safety infrastructure for universities, offices, and public transport networks worldwide
🤝 Emergency services API integration — direct connection to local emergency response systems in supported cities

"Every woman — regardless of where she lives, what language she speaks, or what time it is — deserves a guardian that is smarter than danger, faster than fear, and always on her side."

Safety should be instant. Not delayed. SheShield makes that real.