README.md

metadata

title: PotholeIQ
emoji: 🛣️
colorFrom: blue
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false

PotholeIQ — Pothole Management on the Box Platform

One workflow from citizen intake to verified repair: AI photo classification (YOLOv8), severity scoring, GIS mapping, supervisor review, crew dispatch, digital closeout, and an auditable before/after report — with Box as the content system of record and Supabase as the data mirror.

Pages

Page	Purpose
/index.html	Citizen submission portal (photo + GPS, AI classification)
/command-center.html	Supervisor Command Center (KPIs, map, approve/dispatch/resolve)
/crew-closeout.html?case=ID	Standalone crew closeout form (locked to one case)
/case-report.html?case=ID	Printable before/after report
/track.html	Citizen case tracking

Configuration (Space secrets → env vars)

Box (required)

Secret	Example / note
BOX_AUTH_MODE	client_credentials
BOX_CLIENT_ID	Box app client id
BOX_CLIENT_SECRET	Box app client secret
BOX_SUBJECT_TYPE	enterprise
BOX_SUBJECT_ID	Box enterprise id
BOX_INTAKE_FOLDER_ID	Intake folder id
BOX_CASE_ROOT_FOLDER_ID	Case root folder id (often same as intake)
BOX_METADATA_SCOPE	enterprise
BOX_METADATA_TEMPLATE_KEY	potholeCase

Supabase (recommended — data mirror + supervisor login)

Secret	Note
SUPABASE_URL	https://<ref>.supabase.co
SUPABASE_SECRET_KEY	service-role key (server-side only)
DASHBOARD_AUTH	true to require supervisor login on dashboard APIs

Optional

Secret	Note
CLOSEOUT_NOTIFY_RECIPIENTS	Comma-separated default recipients for before/after reports
BOX_AUTOMATE_SECRET	Shared secret for Box Automate → /api/box/automate-update
DATA_DIR	Set to /data/potholeiq if HF persistent storage is enabled
BOX_CASES_TTL_MS	Box case-list cache TTL (default 45000)

Architecture notes

• Box = source of truth for content: photos, sidecars, generated documents, before/after PDFs (in Closeout Reports), metadata (potholeCase template).
• Supabase = data mirror (db/schema.sql, run once in the Supabase SQL editor).
• The local data/ store is a working cache only — safe to lose on restarts.
• First AI classification after a cold start takes ~30–60 s (model load).