StreetSmart

Inspiration

This project started with something simple and real. We’ve all had that moment hitting a pothole you didn’t see, sliding slightly on black ice, or suddenly hearing sirens and not knowing where the ambulance is coming from. It’s stressful, and in those few seconds, you’re forced to react fast.

We realized that drivers don’t need more screens or more distractions they need smarter support. Something that works quietly in the background and speaks up only when it matters.

Street Smart was born from our own experiences on the road and a shared belief that safety shouldn’t require extra effort. We wanted to build something that protects drivers, supports emergency responders, and strengthens communities all while keeping hands on the wheel and eyes forward.

What It Does

Alerts drivers when an emergency vehicle is approaching (even from behind)
Warns about slippery or hazardous road conditions
Notifies drivers about reported potholes
Allows hands-free pothole reporting using voice
Automatically reports potholes to county repair systems using live location data

How We Built It

Voice System

ElevenLabs API for ultra-realistic Text-to-Speech, giving Navi a natural, conversational voice
Web Speech API for browser-based Speech-to-Text to capture driver voice commands hands-free

AI Processing

Gemini API for natural language understanding and intelligent response generation
- Interprets prompts like “How are the roads?” or “Report a pothole.”
- Determines when to trigger hazard alerts
- Structures responses in a safe, concise, driving-friendly format

Mapping & Location

Google Maps API for:
- Live route tracking
- Map visualization
- Location-based hazard overlays
Browser Geolocation API for real-time GPS coordinates
Distance calculations to detect proximity to hazards and emergency vehicles

System Flow

Driver Voice → Speech-to-Text → Gemini API (Intent Processing)
→ Location + Hazard Matching (Google Maps API)
→ Alert Generation → ElevenLabs Voice Output

Challenges We Ran Into

One of our biggest challenges was making the voice interaction feel natural and reliable. Wake-word activation and speech recognition can easily misfire, especially in a driving environment with background noise.

Another challenge was reducing latency. Since we were using the Gemini API for intent understanding and ElevenLabs for voice output, we had to optimize the pipeline so responses felt instant and not delayed.

Integrating real-time location data from the Google Maps API required careful coordination — we had to ensure hazard detection was accurate and alerts triggered at the right moment, not too early and not too late.

Finally, designing alerts that were helpful but not distracting was harder than expected. Safety means clarity, not overload.

Accomplishments That We’re Proud Of

Built a fully hands-free AI assistant that feels conversational and usable
Navi interprets natural driver prompts in real time
Emergency vehicle alerts work based on proximity detection
Potholes are reported automatically using live GPS data
Voice responses sound realistic and human-like
Successfully connected AI processing, mapping, and speech into one seamless system

Most importantly, we turned a safety idea into a working prototype within a hackathon timeframe.

What We Learned

We learned that voice UX design is just as important as backend logic. The wording, tone, and timing of alerts matter significantly when someone is driving.

We also learned how powerful AI becomes when combined with real-world context like GPS data. Large language models aren’t just chat tools they can power real-time decision systems.

Finally, we learned that building something safety-focused requires constant thinking from the user’s perspective: