BeeSafe

Advertisment
Drive Session with Front Camera Expanded
Drive Session with Face Camera Expanded
Sleep/Fatigue AI Assistance
Accident Detection & Assistance
Gamified Learning and Encouragement
Blockchain Verification of Footage to protect against AI doctoring

Inspiration

Every day, thousands of Australians get into their cars expecting a normal drive home. 🚗

For some, that drive never ends the way they expected.

Last year there were over 1,500 road deaths in Australia.
Driver fatigue contributes to up to 35% of fatal crashes, and driver distraction plays a role in 3 in 5 serious accidents.

A split second of lost attention can mean the difference between arriving safely or becoming another statistic. 😴📱⚠️

But the danger doesn’t end at the moment of the crash. ⚠️

In many accidents, the critical difference between life and death is time. When drivers are unconscious, injured, or in shock, calling for help isn’t always possible. Even a delay of a few seconds can dramatically impact emergency response and survival. 🚑

At the same time, accidents often lead to disputes, scams, and fraudulent claims. Without clear evidence of what actually happened, innocent drivers can be blamed, insurance claims can be contested, and staged accidents can cost victims thousands. ⚖️💰

Dashcams were supposed to solve this, but they come with problems of their own.

Most people can’t afford them, can’t figure out how to install them, or simply don’t have them when they need them most. 📹❌

And even when footage exists, videos can be edited, manipulated, or disputed, leaving drivers without trustworthy proof. 🧾❓

We realized something powerful:

Almost every driver already carries a smartphone. 📱

What it does

BeeSafe turns any smartphone into an AI-powered driving safety platform, combining computer vision, automatic emergency response, and tamper-proof incident recording — without the need for expensive dashcam hardware. 📱🚗

Our system focuses on three core roles:

🔍 The Overseer: Real-Time Driver & Road Awareness

BeeSafe monitors both the driver and the road using real-time computer vision.

Facial landmark detection identifies drowsiness and distraction, triggering an AI voice alert if attention drops. The road-facing camera detects vehicles and hazards, calculating safe following distance and dangerous conditions.

If repeated fatigue is detected, BeeSafe can suggest nearby safe places to pull over using live geolocation.

🛡️ The Protector: Instant Response When Seconds Matter

Using motion sensors and G-force calculations, BeeSafe detects harsh braking and crashes.

If a serious crash occurs, the system automatically:

Calls 000 emergency services
Sends the driver’s exact GPS location
Generates an emergency voice message using AWS Polly to speak to emergency services

Footage is also cryptographically verified on blockchain, protecting drivers from tampered evidence and insurance fraud. ⚖️

📚 The Teacher: Helping Drivers Improve

BeeSafe encourages safer driving through insight and gamification.

Each trip generates a Drive Report logging events like fatigue, distraction, harsh braking, and crashes. Drivers earn a Driver Score and climb leaderboards based on safe behaviour.

Important events can also be replayed and downloaded for review.

How we built it

🔹 Frontend: Built with TanStack Start (React 19). Styled with Tailwind CSS and ShadCN UI, with Framer Motion powering animations.

🔹 Driver Monitoring AI: Real-time driver detection powered by Google MediaPipe FaceLandmarker, tracking 468 facial landmarks to calculate eye-closure and attention metrics. GPU-accelerated inference allows drowsiness detection with sub-33ms latency.

🔹 Traffic Detection: Road awareness implemented using a YOLOv8 object detection model running in the browser with ONNX Runtime Web and WebGPU acceleration, detecting vehicles and hazards in real time. DeviceMotionEvents for G-Force calculations.

🔹 Backend & APIs: Built with a type-safe RPC architecture using oRPC and Zod validation, ensuring end-to-end type safety between frontend and backend services.

🔹 Voice & Emergency Automation: AI voice alerts and emergency call messages generated using Amazon Polly Neural Text-to-Speech for natural, low-latency audio feedback.

🔹 Blockchain Verification: Dashcam footage is cryptographically verified on the Base Sepolia network by embedding SHA-256 video hashes into blockchain transactions, creating tamper-proof proof-of-existence.

🔹 Storage & Database: Video clips are securely uploaded to AWS S3 using pre-signed URLs, while user data, drive reports, and events are stored in PostgreSQL via Prisma ORM.

🔹 Deployment & Infrastructure: Deployed on Vercel. CI/CD is handled through GitHub Actions, while Docker ensures consistent builds and reproducible environments across development and deployment.

Challenges we ran into

🔹 Hackathon Deployment on iOS: Getting a mobile app into judges’ hands during a hackathon is harder than it sounds. Native apps require App Store review or TestFlight approval, which can take time. We built BeeSafe as a web-native platform, allowing judges to simply open a link and run the app instantly on their phones.

🔹 Crash Detection Physics: Detecting crashes reliably was a lot harder than expected. We had to whiteboard the physics and math behind G-force calculations to distinguish between normal driving behaviour, harsh braking, and actual collisions. Finding the right balance between sensitivity and false positives took multiple iterations and a few roadtrips in the car.

Accomplishments that we're proud of

🔹 Real-time AI running on a smartphone: We successfully ran driver monitoring and road hazard detection directly in the browser, achieving real-time performance on a mobile device without requiring specialized hardware.

🔹 Automatic emergency response system: We built a crash detection pipeline that can identify high G-force impacts and automatically initiate an emergency call with precise location data and AI-generated voice communication.

🔹 End-to-end safety ecosystem: BeeSafe isn’t just a dashcam — it combines driver monitoring, hazard detection, crash response, and post-drive analysis into one unified platform.

🔹 Tamper-proof evidence system: By integrating blockchain hashing for video verification, we created a way to prove that recorded footage hasn’t been altered, helping protect drivers from insurance fraud and disputes.

🔹 Web-native accessibility: We delivered the entire experience as a web app, allowing anyone to instantly turn their smartphone into an AI-powered driving safety system with no installation or hardware required.

What we learned

🔹 Mobile-first engineering is hard: Running real-time AI and computer vision in a smartphone browser requires careful optimization, efficient models, and constant attention to performance.

🔹 Web apps can do more than expected: Modern web technologies now allow complex capabilities like real-time vision processing, geolocation, and sensor data directly in the browser.

🔹 Reliable safety systems require iteration: Building crash detection and driver monitoring meant testing assumptions, tuning thresholds, and validating our approach to avoid false positives.

What's next for BeeSafe

🔹 V2V (Vehicle-to-Vehicle) Protocol: Expanding the "Digital Flare" to include road hazard reporting (potholes, debris) verified by AI.

🔹 App-Native: Moving from browsers to native OS apps to improve accessibility of core smart phone features

🔹 Insurance and safety partnerships: Allow users to optionally share verified driving data and tamper-proof incident footage with insurers or authorities to simplify claims and improve trust.