BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pyconde-pydata-2025//speaker//8MXU3M
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-pyconde-pydata-2025-XLZQFA@pretalx.com
DTSTART;TZID=CET:20250424T165500
DTEND;TZID=CET:20250424T174000
DESCRIPTION:Retrieval-Augmented Generation (RAG) chatbots are a key use cas
 e of GenAI in organizations\, allowing users to conveniently access and qu
 ery internal company data. A first RAG prototype can often be created in a
  matter of days. But why are the majority of prototypes still in the pilot
  stage? [\\[1\\]](https://www2.deloitte.com/content/dam/Deloitte/us/Docume
 nts/consulting/us-state-of-gen-ai-q3.pdf)\n\nIn this talk we share our ins
 ights from developing a production-grade chatbot at Merck. Our RAG chatbot
  for R&D experts accesses over 50\,000 documents across numerous SharePoin
 t sites and other sources. We identified three technical key success facto
 rs:\n1. Building a robust data pipeline that syncs documents from source s
 ystems and that handles enterprise features such as replicating user permi
 ssions. \n2. Developing a chatbot workflow from user question to answer wi
 th retrieval components such as hybrid search and reranking\n3. Establishi
 ng a comprehensive evaluation framework with a clear optimization metric.\
 n\nWe think that many of these lessons are broadly applicable to RAG chatb
 ots\, making this talk valuable for practitioners aiming to implement GenA
 I solutions in business contexts.
DTSTAMP:20260422T091121Z
LOCATION:Titanium3
SUMMARY:Lessons learned in bringing a RAG chatbot with access to 50k+ diver
 se documents to production - Bernhard Schäfer\, Nico Mohr
URL:https://pretalx.com/pyconde-pydata-2025/talk/XLZQFA/
END:VEVENT
END:VCALENDAR