FindSafety

Dashboard

Inspiration

South Africa has one of the highest crime rates in the world — something most citizens experience firsthand. Yet, we realized there's no single platform that combines real-time data, AI insights, and community involvement to help people stay safe.

We searched for forums or community-driven platforms where people could warn each other, like Waze does for traffic. Surprisingly, nothing like that existed here. That’s when we knew we had to build FindSafety — a powerful, AI-driven, community-first crime awareness tool.

What it does

FindSafety is a smart crime awareness platform designed for South African communities. Here’s what it offers:

AI-Powered Chat: Ask questions like “Is it safe in Joburg CBD?” — the bot fetches insights from real crime data.
Semantic Crime Search: Find incidents with natural language using vector search on crime descriptions.
Community Alerts: Users can post warnings, sightings, or tips. Think Waze, but for safety.
Firebase Notifications: Get notified when crimes are reported nearby.
Official Reporting Tool: Offers contact details for nearby police if users want to report something officially.
Smart Tips Generator: The bot posts auto-generated safety advice based on recent crime trends.
Filter by Type, Date, Location, Severity: For power users needing specific search tools.

How we built it

Backend: Flask (Python), MongoDB (collections: crimes, alerts, raw_scraped_data)
Frontend: HTML, CSS, TypeScript (mobile-first, with plans for PWA)
AI: Google Gemini API (for chat and vector embedding)
Vector Search: MongoDB Atlas vector indexing on description_vector
Notifications: Firebase Cloud Messaging
Hosting: Ready for Google Cloud deployment

We also scraped publicly available crime-related content and used embeddings to match AI-generated queries to real-world reports.

Challenges we ran into

Lack of structured crime data: We had to scrape, clean, and normalize unstructured content (e.g., Facebook posts).
AI grounding: Ensuring the chatbot gives useful, fact-based responses without hallucinations.
No existing forums: We planned to scrape local reports but had to build our own platform instead!
Time vs. Scope: Balancing ambitious features (like police station integration) with tight deadlines.

Accomplishments that we're proud of

Created a working crime chatbot grounded in real data and vector search.
Designed a community alert system that makes real-time reporting possible.
Integrated Firebase for contextual notifications.
Auto-generated safety tips tailored to current crime trends.
Designed a system that can scale beyond a demo — with real impact potential.

What we learned

How to use vector embeddings and semantic search for natural interaction with structured crime data.
The importance of designing for trust and safety when building public tools.
Just how big the gap is in South Africa’s community crime tech space — and how important this project could be.

What's next for FindSafety

Launch as a Progressive Web App for wider mobile accessibility.
Add live location-based alerts and heatmaps of crime activity.
Partner with SAPS or NGOs to offer verified reporting and victim support.
Integrate machine learning to forecast crime risks in neighborhoods.
Translate the app into multiple South African languages.
Grow a nationwide community-driven safety network — because safety should be shared.