SafeSteps AI

Navigate Toronto's streets safely using real-time crime data, AI analysis, and intelligent routing algorithms

Comment

SafeSteps AI: The Safest Way Home in Toronto

About the Project

Every year Toronto records more than 70,000 crimes — assaults, robberies, car thefts, and violent incidents. If you’ve lived here for years, you already have that “street instinct”: you know which corner to avoid after dark, which alley feels wrong, which path is safe.

But millions of people don’t have that instinct yet:

  • International students arriving at 11 PM with suitcases
  • Tourists stepping out of Union Station
  • Newcomer families with kids
  • Uber and delivery drivers working late in unknown neighborhoods

For them, one wrong turn can be dangerous.
SafeSteps AI fixes that. We turn public crime data into a real-time, street-by-street safety map so everyone — from day one — can walk, bike, or drive the safest possible route.

What Inspired Me

I saw too many friends and classmates feel scared just because they didn’t know the city yet. Safety shouldn’t require years of experience. It should be on the map.

How I Built It

  1. Data Sources

    • 10+ years of Toronto Police Public Safety Data (158 neighbourhoods)
    • OpenStreetMap full topology (11,000+ intersections, 13,000+ street segments, POIs, lighting, etc.)

  2. Geospatial Processing (Python)

    • Point-in-Polygon: every intersection inherits its neighbourhood’s crime rate
    • Feature engineering:
      • Density of bars/clubs in 100 m radius
      • Street type & hierarchy
      • Intersection complexity (number of connected roads)
    • Final risk score (0–100) per intersection using weighted formula
    • Street segments = average of their two endpoints

  3. Graph & Routing

    • Built weighted graph with NetworkX
    • A* algorithm finds the safest path (not the shortest)

  4. Web App

    • Leaflet.js dark-mode interactive map
    • Streets colored green / yellow / red
    • Real-time route calculation
    • Click any street → see exact risk score and reasons

  5. Live Updates

    • Google Gemini API reads Toronto Police press releases and news daily
    • Automatically adjusts risk weights when new incidents are reported

Challenges I Faced

  • Merging neighbourhood-level crime data with street-level geometry (solved with spatial indexing)
  • Keeping the app fast with 11,000+ nodes (vectorized calculations + caching)
  • Making sure the map informs without stigmatizing communities (transparent methodology + focus on infrastructure factors)

Built With

  • Python
  • GeoPandas, Shapely, NetworkX
  • Leaflet.js
  • Google Gemini API
  • OpenStreetMap
  • Toronto Police Open Data
  • Vercel (hosting)

Try It Out

Project Media

(Upload these 4 images to Devpost – they look professional on dark background)

  1. Data fusion & point-in-polygon diagram
  2. Risk score formula with weights
  3. Shortest vs safest route example (Union Station → TMU)
  4. Screenshot of the live interactive map

Video Demo (90 seconds) → https://youtu.be/M0kzqOYuxMA

SafeRoute AI is more than an app — it’s a step toward UN SDG 11 (Sustainable Cities), SDG 9 (Innovation), and SDG 16 (Peace & Justice).
Let’s make Toronto the first big city where safety is truly available to everyone, from day one.

Built With

Share this project:

Updates