BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//foss4gnl//talk//VB9UMZ
BEGIN:VTIMEZONE
TZID:Europe/Amsterdam
BEGIN:DAYLIGHT
DTSTART:20250709T000000
TZNAME:CEST
TZOFFSETFROM:+0200
TZOFFSETTO:+0200
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20251026T030000
RDATE:20261025T030000
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20260329T030000
RDATE:20270328T030000
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:From polygons to AI: defensive patterns from a five-city PPGIS pip
 eline - Mahdi Rahimi
DTSTART;TZID=Europe/Amsterdam:20260709T110000
DTEND;TZID=Europe/Amsterdam:20260709T112000
DTSTAMP:20260726T041152Z
UID:pretalx-foss4gnl-VB9UMZ@pretalx.com
DESCRIPTION:AccessCity4All is a research project comparing objective\, per
 ceived\, and actualized accessibility across five EU cities. Within the pr
 oject\, WP5 collects PPGIS data via Maptionnaire: respondents draw their p
 erceived neighborhood on a map and mark home\, work\, and activity locatio
 ns as points. This talk is about the R pipeline that turns those surveys i
 nto reproducible analytic datasets.\nBelow the survey layer everything is 
 open: R + sf for geometry\, r5r / OSRM / OpenRouteService for routing\, Ge
 oPackage and GeoJSON for interchange\, and a Mistral-driven step for multi
 lingual free-text reclassification.\nWhat it actually taught us is what ha
 ppens when around 5\,000 respondents across five cities draw their own nei
 ghborhoods\, around 18\,000 trips need real routed distances\, and free-te
 xt responses arrive in five languages. This talk walks through three defen
 sive patterns the codebase grew over five years:\n\n- a geometry-repair pi
 peline that turns freehand citizen drawings into valid analytic polygons\,
  with home-inside verification that catches its own cache mistakes\;\n- a 
 dual-mode activity-space classifier built on routed distances and pure-R c
 lustering\, with a Euclidean fallback when the routing engine is flaky\;\n
 - an audit-logged AI workflow that's defensible in peer review.\n\nIf you 
 write geospatial R that has to survive contact with real data\, this talk 
 is for you.
LOCATION:Brunsemazaal
URL:https://pretalx.com/foss4gnl/talk/VB9UMZ/
END:VEVENT
END:VCALENDAR