BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//compute-paris-2026//talk//9HEGSQ
BEGIN:VTIMEZONE
TZID:Europe/Paris
BEGIN:STANDARD
DTSTART:20251126T000000
TZNAME:CET
TZOFFSETFROM:+0100
TZOFFSETTO:+0100
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20260329T030000
RDATE:20270328T030000
TZNAME:CEST
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:20261025T030000
RDATE:20271031T030000
TZNAME:CET
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
SUMMARY:Batch processing for constrained generation - Alonso Silva
DTSTART;TZID=Europe/Paris:20261126T120000
DTEND;TZID=Europe/Paris:20261126T123000
DTSTAMP:20260930T121004Z
UID:pretalx-compute-paris-2026-9HEGSQ@pretalx.com
DESCRIPTION:Processing large volumes of unstructured or semi-structured da
 ta into analysis-ready dataframes remains a core challenge in modern data 
 pipelines. Structured outputs (leveraging the function/tool-calling capabi
 lities of LLMs) and structured generation (which guarantees 100% valid JSO
 N by constraining token sampling to a predefined schema) offer powerful\, 
 complementary solutions to this challenge.\n\nIn this talk\, we introduce 
 both methods\, discuss their respective advantages and limitations in the 
 context of batch data processing\, and demonstrate why combining both appr
 oaches yields superior results compared to either method alone. To illustr
 ate these concepts\, we present real-world applications including:\n\nBatc
 h data extraction from unstructured text into structured dataframe columns
 \nLarge-scale text classification for categorical feature engineering\nEnt
 ity and relationship extraction to enrich dataframes for downstream analyt
 ics\nA live demonstration will showcase an end-to-end pipeline processing 
 real datasets\, highlighting throughput\, schema compliance\, and integrat
 ion with standard dataframe libraries (e.g.\, pandas\, Polars).
LOCATION:Room 106
URL:https://pretalx.com/compute-paris-2026/talk/9HEGSQ/
END:VEVENT
END:VCALENDAR