Your road companion · ShellHacks 2026

A calmer way
to get there.

Know more before you go. Stay aware along the way. WheelBuddy pairs route insights with real-time driving alerts, right on your iPhone.

WheelBuddy began at ShellHacks 2026. Encouraged by our live tests, we’re refining the app with the goal of making it available to the public.
◈BEFORE THE DRIVECompare your routes
WheelBuddy app landing screen
◉ON THE ROADStay in the know
25°45′ N   80°11′ W
Scroll to explore01 / 05
01WHY WE BUILT IT

It started with
a drive.

On a trip to the Smoky Mountains, we rented a car with modern driver assistance. We felt more aware and reassured behind the wheel. Back home in our older cars, we wondered whether the phone we already owned could help fill some of that gap.

We also wanted more context before setting out, especially after dark. The fastest route isn’t always the one we feel most comfortable taking.

So we built a companion for both decisions: choosing a route and driving it.

THE RESEARCH THAT INSPIRED US

Studies by IIHS associated forward collision warning with 27% fewer rear-end striking crash involvements and lane departure warning with 11% fewer single-vehicle, sideswipe and head-on crashes after adjustment. These are findings for the studied vehicle systems, not measured outcomes for WheelBuddy.

02THE GOAL

An AI companion
for your next journey.

From choosing a safer route to navigating the road ahead, WheelBuddy helps you stay informed and aware throughout the drive.

SAFETY FIRST

Choose safer routes

WheelBuddy predicts estimates crime risks associated with all available routes to your destination and allows you to choose routes that are safer. Compare routes using estimated crime risk for all available areas and time of travel.

  • Route-by-route safety estimates
  • Risk broken down by crime category
  • Hourly safety insights
Watch navigation test ↗
A
B
✦ROUTE INSIGHTAnalyze safety of available routes
Illustrative route visual · app estimates vary by trip
03BUILT FOR THE WHOLE DRIVE

Plan with insight
Drive with awareness

Practical signals at the moments they matter, designed to be easy to understand at a glance.

01 / ROUTES

Route safety
insights

See estimated crime risk alongside travel choices, with category and hourly views that add context to the journey ahead.

02 / ROAD

Road awareness

On-device object detection watches for nearby vehicles and pedestrians.

03 / DRIVER

Attention alerts

Face orientation helps identify when the driver looks away for too long.

04 / DRIVER

Drowsiness cues

A temporal model detects prolonged eye closure and triggers audio and visual warnings.

05 / ROAD

Lane assistance

On-device lane detection explores another way to notice drift. This feature is still being refined.

06 / DRIVE

Speed monitor

GPS shows your current speed in Drive Mode. During active navigation, Google Maps can supply an estimated posted limit, color cues when you’re over it, and a “Slow down” banner for major overspeed.

04OUT IN THE REAL WORLD

See it in motion.

Clips from our live tests. Select a feature to watch how WheelBuddy responds in real-world scenarios.

Prototype demonstrations. Alerts and predictions are assistive information and do not replace attentive driving.

05 BEHIND THE BUILD

How WheelBuddy works

WheelBuddy connects time-aware route predictions with live road and driver monitoring. FastAPI serves the crime and drowsiness models, while Core ML and Apple Vision process camera signals on the iPhone.

01 / CRIME PREDICTION DATA

Crime model and coverage

We trained our crime prediction model using outdoor crime records from the Crime Open Database (CODE). The selected data covers records from 2016 to 2025 across 11 cities:

  • Boston
  • Charlotte
  • Detroit
  • Louisville
  • New York
  • Virginia Beach
  • Austin
  • Fort Worth
  • Houston
  • Memphis
  • Nashville
02 / ROUTE SAFETY ANALYSIS

Time-aware route scoring

Google Maps supplies candidate routes and travel times. We identify the areas each route crosses and when the driver is expected to reach them, then request predictions from our crime model through FastAPI.

The app combines those predictions into route safety scores, crime category breakdowns, and hourly views.

Route and ETA from Google Maps API Per-route area (as H3 cells) extraction Risk estimate using crime prediction model Per-route safety scoring
03 / ON-DEVICE VISION

Apple Core ML and AppleVision API

The iPhone runs its front and rear cameras simultaneously. Core ML identifies vehicles and pedestrians in road footage, while Apple Vision estimates the driver's head orientation for looking-away alerts.

A ResNet-18-based lane model trained on the CurvedLanes dataset also runs through Core ML. Improving lane detection across roads and lighting conditions remains an active focus.

04 / DROWSINESS

Drowsiness detection

We trained a custom temporal model using the Driver Monitoring Dataset (DMD). WheelBuddy extracts eye and facial measurements over time and sends them to the model through an inference API.

Sustained eye closure triggers visual and audio warnings, bringing the model's output into the live driving experience.

WHERE WE GO FROM HERE

Good journeys
are never finished.

Our real-world tests have been encouraging. Next, we’re improving lane detection across more roads and lighting conditions. Additionally, for stronger inattention detection, want to use gaze direction estimation instead of head orientation.

With broader testing and refinement, our goal is to bring WheelBuddy to the App Store.

Explore the project on GitHub
Live test