BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//speaker//L78HGT
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-CBE9UC@pretalx.com
DTSTART;TZID=CET:20260911T132500
DTEND;TZID=CET:20260911T135500
DESCRIPTION:Standard MLOps stacks assume you are operating tens or hundreds
  of models. What happens when you need to train\, validate\, and serve tho
 usands of forecasting models on a weekly cadence\, producing millions of f
 orecasts for a national energy portfolio? Tools like MLflow start to buckl
 e\, per-model evaluation stops being meaningful\, and the choice between o
 ne global model and many local ones becomes an architectural decision rath
 er than a modeling one.\nThis talk is a systems architecture case study of
  the platform we built when off-the-shelf MLOps stopped working. We focus 
 on three concrete\, transferable patterns:\n\n1. Heterogeneous parallel tr
 aining — running Spark ML global models and thousands of local models (v
 ia applyInPandas) in a single unified workflow\, with a four-axis decision
  framework (data volume per entity\, signal heterogeneity\, cold-start beh
 avior\, operational cost) for routing each segment.\n2. A model registry t
 hat scales beyond MLflow — what we kept\, what we dropped\, and the sche
 ma that lets us manage versions\, parameters\, and artifacts for a vast po
 rtfolio without metadata-store collapse.\n3. Portfolio-level promotion gat
 es — why per-model accuracy metrics mislead at scale\, and how we run ch
 ampion/challenger experiments with strict temporal integrity and aggregate
  decision criteria.\n\nAttendees will leave with patterns they can apply t
 o any large-scale time-series system\, not just energy forecasting.
DTSTAMP:20260710T150511Z
LOCATION:Room 1 (170)
SUMMARY:Systems for Scale: Architecting a Nationwide Energy Forecasting Pla
 tform - Mohit
URL:https://pretalx.com/pydata-amsterdam2026/talk/CBE9UC/
END:VEVENT
END:VCALENDAR