SeeAhead

Inspiration

1 in 5 people worldwide experience some form of neurodivergence or social anxiety. We watched friends and family struggle in everyday situations — a busy MRT station, a new workplace, a social gathering — not because they lacked intelligence or empathy, but because the unwritten rules of the world aren't written anywhere. We wanted to change that.

What it does

SeeAhead is an AI-powered social navigation app. Point your camera at any scene and get an instant, calm, plain-language breakdown of what's happening — the mood of the space, what people around you are feeling, what to expect next, and any sensory challenges to watch out for. When things feel overwhelming, SeeAhead checks in, guides you through a breathing exercise, and puts the people you trust just one tap away.

How we built it

• Frontend: React Native + Expo SDK 55 with expo-camera for real-time capture
• Backend: Node.js + Express handling multipart image uploads via Multer
• Vision AI: Reka Flash (reka-flash) — Reka's multimodal vision model analyzes each scene and returns structured, autism-friendly guidance
• Pipeline: Image → base64 → Reka API → structured 4-section analysis → displayed on device in under 3 seconds
• Smart monitoring: Frame fingerprinting detects meaningful scene changes (18%+ pixel diff) to avoid redundant API calls during continuous watching mode

Challenges we ran into

• Getting Reka's vision model to consistently output our structured 4-section format required careful prompt engineering
• Connecting a physical phone to a localhost backend required solving CORS, correct LAN IP detection, and Windows firewall rules
• Balancing response speed with analysis quality — we tuned image quality settings to hit under 3s round-trip on a local network
• Designing for the actual user: tone matters enormously. A response that feels condescending or uses idioms is worse than no response at all

Accomplishments that we're proud of

• Full end-to-end pipeline working on a physical device in under 48 hours
• Reka vision model accurately identifying social cues, body language, and sensory risks in real-world photos
• A genuinely calm, literal, non-condescending AI voice that we'd be comfortable putting in the hands of an autistic teenager
• Smart scen ## What we learned
• Multimodal vision models like Reka are remarkably capable at reading social context — far beyond what we expected
• Prompt design for accessibility is a discipline of its own. The difference between a helpful and harmful response is often a single sentence
• Building for a specific user group forces you to make better product decisions across the board — clarity, simplicity, and calm design benefit everyone

What's next for SeeAhead

• Voice readout — read analysis aloud for hands-free use
• Customizable contacts — let users add their own trusted people
• Wearable integration — detect stress passively via heart rate, trigger check-ins automatically
• Journey analytics — map the places that feel safe vs. overwhelming over time
• Offline mode — on-device model for privacy and zero-latency response

Built With

• expo.io
• express.js
• node.js
• react
• react-native
• reka