BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//talk//DAUQVR
BEGIN:VTIMEZONE
TZID:CET
BEGIN:STANDARD
DTSTART:20001029T040000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000326T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-pydata-amsterdam2026-DAUQVR@pretalx.com
DTSTART;TZID=CET:20260911T100500
DTEND;TZID=CET:20260911T105000
DESCRIPTION:Serving user-specific ML models at scale while satisfying low-l
 atency requirements is non-trivial\, especially when the model training co
 de is regularly updated\, making models trained weeks apart require potent
 ially incompatible environments. How do you build a scalable inference pip
 eline with per-user models while still moving fast\, when the runtime keep
 s evolving? In this talk\, I’ll show how our small team built and mainta
 ins a Python inference pipeline that can serve thousands of user-specific 
 models from a single service using a dynamic model loading approach. There
  is still one challenge to tackle: enabling training code and dependency u
 pdate without risking incompatible inference environments. We’ll compare
  different strategies and explain the solution we chose to stay agile whil
 e preserving a smooth UX and DevEx. Finally\, we will walk through the end
 -to-end architecture built using open source tools such as MLflow\, MLServ
 er\, KServe\, and supporting both local development and production-scale d
 eployments.
DTSTAMP:20260710T141133Z
LOCATION:Room 2 (350)
SUMMARY:Serving Personalized ML at Scale with Evolving Runtimes - Rayan Dao
 d
URL:https://pretalx.com/pydata-amsterdam2026/talk/DAUQVR/
END:VEVENT
END:VCALENDAR