BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.pydata.org//pydataglobal2025//talk//AJD8TU
BEGIN:VEVENT
UID:pretalx-pydataglobal2025-AJD8TU@cfp.pydata.org
DTSTART:20251211T113000Z
DTEND:20251211T120000Z
DESCRIPTION:In this presentation\, we demonstrate how Large Language Models
  (LLMs) can revolutionize safety log analysis in the oil and gas industry.
  Our research with a major operator involved processing 15\,000 safety obs
 ervations through a novel multi-stage pipeline. First\, we developed a dom
 ain-specific categorical framework aligned with industry standards. We the
 n implemented an unsupervised learning approach using sentence transformer
 s to calculate semantic similarity between observations and predefined cat
 egories. This enabled multi-dimensional classification with weighted confi
 dence percentages. Finally\, we deployed a fine-tuned LLM to assign priori
 ty scores and enhance categorization accuracy\, all while maintaining data
  privacy through on-premises processing. The resulting system streamlines 
 real-time safety log processing\, enabling more efficient identification o
 f potential hazards and trends. Our implementation demonstrates significan
 t improvements in classification accuracy and processing efficiency compar
 ed to traditional methods\, providing actionable insights for proactive sa
 fety management.
DTSTAMP:20260518T180206Z
LOCATION:Machine Learning & AI
SUMMARY:Revolutionizing Safety Log Analysis in Oil and Gas: A Multi-Stage L
 LM Approach for Enhanced Hazard Identification - Andrew Yule\, Iain Docher
 ty
URL:https://cfp.pydata.org/pydataglobal2025/talk/AJD8TU/
END:VEVENT
END:VCALENDAR