BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pydata-amsterdam2026//speaker//ETDLD3
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-Y7CCXA@pretalx.com
DTSTART;TZID=CET:20260911T145500
DTEND;TZID=CET:20260911T152500
DESCRIPTION:Search works great when you know what you're looking for. But w
 hat happens when a user types "sunset tours in Barcelona for a couple unde
 r 50 euros"? Traditional search pipelines break down.\n\nWe built an AI-po
 wered discovery agent that lets users describe what they want in natural l
 anguage and get relevant\, ranked recommendations. In this talk\, we'll wa
 lk through the production architecture: a multi-stage LangGraph pipeline t
 hat chains intent interpretation\, parallel retrieval\, LLM-based relevanc
 e filtering\, and natural language synthesis.\n\nWe'll cover the practical
  trade-offs we made along the way\, including a dual-LLM tier strategy (fa
 st nano model for high-volume filtering\, full model for nuanced interpret
 ation) that keeps latency and cost in check without sacrificing quality.\n
 \nWe'll also share how we built evaluation infrastructure with 800+ determ
 inistic tests and zero LLM-as-judge in CI\, using a pass@k approach with p
 ytest to handle output variability while keeping the test suite fast and r
 eliable. You'll walk away with concrete patterns for shipping LLM-powered 
 features to production: model tiering\, deterministic evals\, and graceful
  degradation by design.
DTSTAMP:20260710T150459Z
LOCATION:Unconference
SUMMARY:From Query to Discovery: Building an AI Agent That Helps Travelers 
 Explore - Giampaolo Casolla\, Steven Mi
URL:https://pretalx.com/pydata-amsterdam2026/talk/Y7CCXA/
END:VEVENT
END:VCALENDAR