BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//talk//ABUNQA
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-ABUNQA@pretalx.com
DTSTART;TZID=CET:20260910T150500
DTEND;TZID=CET:20260910T153500
DESCRIPTION:In online marketplaces\, bad actors utilize automation to scale
  fraud quickly.  ML models can catch it\, but they need features computed 
 based on user behavior — updated in a timely manner. Batch features are 
 too slow — fraud spreads before the next scheduled job runs.\n\nDrawing 
 from production experience operating 25+ streaming jobs for a large online
  second-hand marketplace\, we will walk through the architecture and key d
 esign decisions behind our streaming platform powering ML fraud detection.
   This is composed of Kafka topics\, Flink jobs\, a feature store serving 
 real-time inference and a data warehouse for historical data for model tra
 ining. We will touch upon the strategies to bridge the gap between raw str
 eaming output and training-ready data.\n\nWe will also explore fundamental
  differences between streaming and batch feature engineering. From bounded
  vs unbounded data to stateful operators\, fault tolerance\, and scaling 
 — we will cover what makes Flink excellent for real-time processing. Whe
 n real-time isn't needed\, Flink can do batch too — it's just probably n
 ot the best choice\, e.g. dbt with your favourite data warehouse can do be
 tter — we will shed some light on why.\n\nThe principles shared will hel
 p you evaluate what to look for in a stream processing framework and wheth
 er you need Flink's capabilities for your own use cases. Aimed at data eng
 ineers\, ML engineers\, and data scientists interested in streaming vs bat
 ch trade-offs\, fraud detection\, or trust & safety.\n\n**You will walk aw
 ay with:**\n- Understanding of a production tested architecture for servin
 g streaming features to both real-time inference and model training withou
 t duplicate computation\n- Appreciate when real-time features add value vs
  when batch is the right tool\n- A sense of the operational challenges —
  to help you evaluate whether real-time is worth the cost and effort for y
 our specific use case
DTSTAMP:20260710T141224Z
LOCATION:Room 2 (350)
SUMMARY:Real-time vs Batch Features for ML: Lessons from Fraud Detection at
  Scale - Csanád Bakos
URL:https://pretalx.com/pydata-amsterdam2026/talk/ABUNQA/
END:VEVENT
END:VCALENDAR