Toronto Ward 11 Accessibility Pilot
CurbClear is an open-source, zero-form sidewalk barrier reporting and routing tool built for wheelchair users, low-vision pedestrians, and local volunteers. Snap one photo with GPS to classify obstructions, deduplicate overlapping reports, coordinate quick neighborhood fixes, or dispatch structured Toronto 311 (Open311) tickets.
1. Pilot Zone & Field Schedule
The initial 30-day pilot focuses strictly on a 1.5 km radius spanning Spadina Ave, College St, the University of Toronto St. George perimeter, and Hospital Row in Ward 11 (University–Rosedale), targeting winter snow windrows at curb cuts, broken tactile indicators, un-ramped construction detours, and malfunctioning Accessible Pedestrian Signals (APS).
- Days 1–3 · Seed Street Audit: 4–5 local volunteers walk core Hospital Row and Spadina/College routes to log 80–100 baseline barriers so the hazard map is immediately actionable on day one.
- Weeks 1–2 · Ground-Truth Validation: Share the live Ward 11 avoidance map with local accessibility communities and transit riders to verify route accuracy, catch blind spots, and tune severity labels.
- Weeks 2–4 · Dual-Track Resolution: Route light obstructions (snow windrows, blocked curb cuts, parked bikes) to nearby volunteers with before/after photo verification; package structural AODA gaps into Open311 requests and Ward 11 councillor summaries.
2. Open Architecture & Schema
Built on Next.js (WCAG 2.2 AA compliant, voice reporting, IndexedDB offline queue for subway and underpass dead zones), FastAPI async vision pipelines, and PostgreSQL 16 with PostGIS and pgvector. To prevent urban GPS drift from merging distinct intersection corners, deduplication combines a 20-meter spatial radius with CLIP image embedding cosine distance.
-- 001_init_hazards.sql (PostGIS + pgvector hybrid dedup)
SELECT id, (image_embedding <=> $4) AS img_dist
FROM accessibility_hazards
WHERE status IN ('open', 'claimed', 'dispatched_311')
AND category = $1
AND ST_DWithin(
location,
ST_SetSRID(ST_MakePoint($2, $3), 4326)::geography,
20
)
AND (image_embedding <=> $4) < 0.25
ORDER BY img_dist ASC LIMIT 1;
Full migration file: 001_init_hazards.sql. Municipal boundary polygons (ST_Contains) and Open311 GeoReport v2 templates are isolated as pluggable city adapters so other Canadian municipalities (Vancouver, Ottawa, Montreal) can deploy with a single GeoJSON swap.
3. Continuous Ops with Devin Bot
CurbClear uses Devin Bot as a persistent 24/7 engineering and operations teammate connected to our project Discord and GitHub repository:
- Shared Project Memory: Retains Toronto Open311 field quirks, AODA severity thresholds, and historical PostGIS/VLM tuning decisions across contributor sessions.
- Discord Field Triage: Volunteers flag misclassified street hazards in Discord; mentioning
@Devinpulls bad cases, opens GitHub Issues, and appends regression samples. - 311 Alert Auto-Remediation & PR Audits: Monitors
#ops-alertsfor Open311 schema or rate-limit failures to open tested fix PRs, and runs Playwright + axe-core WCAG 2.2 AA checks on incoming GitHub PR webhooks. - Weekly Ward 11 Digest: Audits 72h-stalled 311 tickets daily and compiles a Friday escalation digest for Ward 11 follow-up.
4. Join the Pilot
We are recruiting local walkers in Downtown Toronto for the late-November seed mapping run, as well as contributors interested in Open311 adapters and WCAG 2.2 testing.
Email support@nextjump.tech with subject Ward 11 Pilot to join the project Discord and receive the mapping checklist.