SafeStep > Navigate home without fear.

Inspiration

Personal stories and community reports about the “last mile” — people (especially women and gender-diverse folks) feeling unsafe walking from transit to home late at night.
Existing navigation apps prioritize speed and willfully ignore contextual safety signals (lighting, crowds, informal community reports).
The rise of AR-capable phones — opportunity to give real-time spatial guidance that’s visible and intuitive without forcing eyes off the environment.
Desire to create a privacy-respecting, verifiable reward (non-transferable badge) that recognizes completed safe journeys and encourages trustworthy reporting.
Hackathon prompt to combine AI + AR + a strong evidence/proof story — perfect fit for a product that’s practical and demonstrable.

Analyzes multiple safety signals (lighting, community reports, historical incidents, crowdsourced observations) to compute a safety score for routes.
Offers two route modes: Safest route (default) and Fastest route (optional)—explicitly showing tradeoffs to the user.
Presents an AR-guided path over the phone camera so users follow a clear, glowing route in the real world.
Alerts users when they step off the safe path with clear microcopy and visual cues.
Optionally mints a Soulbound Safe Passage badge (testnet) after a verified safe arrival — a verifiable proof artifact for social / civic use.
Includes an accessible fallback (text-based checkpoint list) for users who cannot or prefer not to use AR.
Privacy-first: location streams used for routing are ephemeral; persistent storage limited to optional hashed proofs for minting and anonymized analytics.

Client (mobile web / PWA + native wrapper) — React Native / Expo for quick dev + WebAR fallback for no-install users.
AR presentation — WebXR / WebAR surface for browser fallback; native ARKit/ARCore when available for best stability.
Mapping & routing — Map tiles & POIs from OpenStreetMap, routing overlay combines map routes + safe-path adjustments.
AI decision layer — Goose (AI agent) mediates multi-signal analysis: lighting estimates, incident summarization, and community report weighting.
Backend — lightweight Node/Express prototype for API, with a toggle to run entirely on data for demos.
Proof & minting — WalletConnect for demo, Polygon testnet for low-cost minting of a soulbound badge; metadata includes route hash, time, and safety score.
Data sources — public datasets (lighting/POI), user-submitted reports, optional local open crime datasets, and real-time device sensors.
Developer & design — Figma Make prompt generated high-fidelity prototype, auto-export assets for Devpost screenshots and a one-page Criteria Evidence summary.

Frontend: React Native (Expo) + WebXR fallback
AR: ARKit / ARCore + three.js overlays or WebXR layers
Maps: OpenStreetMap + Mapbox-style tiles (or MapLibre)
AI orchestration: Goose agent
Backend: Node.js + Express
Wallet / blockchain: WalletConnect + Polygon testnet (SBT)
Design & prototype: Figma (Make prompt for rapid generation)
Testing: Manual QA, device smoke tests across iPhone and Android, accessibility linting

Data sparsity for lighting — public lighting datasets are inconsistent across cities; required synthesizing lighting from POIs, satellite imagery proxies, and community reports.
Balancing safety vs. surveillance — using police incident data can disproportionately affect communities; we had to design weighting and explainability to mitigate bias.
AR stability in low-light — AR tracking is less robust at night; needed fallback UX when camera tracking dropped or drifted.
Third-party services — for the hackathon we had to convincingly Goose and wallet flows while keeping demo believable.
Privacy tradeoffs — proving a route without storing raw GPS required building a verifiable hashing approach while staying simple for demo judges.
Time constraints — polishing the handoff artifacts (developer tokens, export images, evidence PDF) while implementing core flows was tight.
Device fragmentation — varying camera permissions and WebXR support across Android and iOS required careful prototype branching and clear permission microcopy.

End-to-end prototype from onboarding → Goose analysis → AR guidance → minting simulation, demo-ready in Figma and on-device.
Clear evidence layer built into the design that maps UI artifacts to judging criteria (Clarity, Proof, Usability, Rigor, Polish).
Privacy-first proof model: optional hashed route proof that’s adequate for mint metadata without storing identifiable traces.
Soulbound mint proof of concept (testnet) that demonstrates how a verifiable, non-transferable badge can incentivize safe behavior and responsible reporting.
Accessible fallback flow for users who cannot use AR — shows we considered inclusion from the start.
Decision log documenting tradeoffs (e.g., community reports weighted higher than raw incident counts to avoid over-policing).
Design system & export: tokens, assets, and demo screenshots generated for Devpost and judges — polished and consistent.

Explainability matters: judges and users want to see why a route is safe. An AI analysis chain (Goose → lighting → crowd reports) dramatically increases trust.
Design for failure modes: AR works great when conditions are right — but you must design graceful fallbacks (text checkpoints, audible guidance).
Bias mitigation is not optional: naive use of police incident data can harm communities; explicit weighting and documented decisions are necessary.
Privacy wins trust: minimizing stored location data and offering opt-in hashed proofs improved user comfort in testing.
Small UX details matter: clear permission microcopy (“SafeStep needs camera + location…”) and large touch targets made our prototype feel production-ready.
Judges appreciate evidence-first submissions: the Criteria Evidence page and a one-page PDF summarizing proof were frequently complimented during quick demo walkthroughs.

Implement real lighting ingestion using municipal open data and satellite-derived illuminance approximations.
Run a small, opt-in pilot with a local transit agency or university campus to collect community reports and improve the model.
Harden AR fallback UX: add audible directions, haptic cues, and a low-bandwidth text-checkpoint flow.
Produce a one-page Criteria Evidence PDF and short demo video (5 storyboard frames) for judges and partners.

Launch a privacy-safe beta: opt-in reporting, hashed route proofs, and audit logs for bias analysis.
Integrate more public safety datasets with clear provenance and versioning.
Add multi-language support and improved accessibility (screen-reader-first flows, reduced-motion defaults).
Explore partnerships with transit agencies, universities, and nonprofits focused on community safety.

Scale the minting model into a trusted civic credential (with careful governance and anti-abuse safeguards).
Build policy and research collaborations to publish findings on safe routing and community reporting impacts.
Commercial considerations: B2B offerings for campuses or events, and grants / sponsorships for public deployments.