BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.com//pyconde-pydata-2026//talk//EWZMJK
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-2026-EWZMJK@pretalx.com
DTSTART;TZID=CET:20260416T101500
DTEND;TZID=CET:20260416T104500
DESCRIPTION:Keywords: **Explainable AI\, enhanced RAG\, GraphRAG\, LLMOps\,
  dialog system evaluation.**\n\nDesigning reliable dialog flows for LLM-ba
 sed systems remains challenging once conversations require branching\, cor
 rection\, or multi-step reasoning. Dialog graphs often evolve organically 
 and accumulate structural issues: endless correction loops\, dead subpaths
 \, redundant validation steps\, overly generic catch-all branches\, or lin
 ear sequences that should be collapsed. Such phenomena raise operational c
 osts\, significantly increase TTFT and make the system answer less predict
 able and explainable.\n\nMany solutions try to introduce an all-fit genera
 lized RAG retrieval solution. Contrary to this\, we present our empirical 
 learnings on how to enhance system speed\, lower overall costs and offer a
  better dialog graph explainability through enhanced LLM call tracing and 
 iterative enhancements for common dialog paths.\n\nWe also show that more 
 elaborated knowledge retrieval strategies like GraphRAG may drastically en
 hance overall response quality and shorten the dialog graph. We evaluate s
 everal approaches and give recommendations on how to leverage more complex
  document indexing phases for inference time benefits.\n\nOverall\, the se
 ssion argues that scalable conversational systems require not only better 
 prompts\, but explicit graph structures paired with rigorous tracing and d
 ata-driven optimization.
DTSTAMP:20260523T180013Z
LOCATION:Europium [3rd Floor]
SUMMARY:Don’t call your LLM too often! How to build your dialog graph wit
 h confidence and sleep at night. - Evgeniya Ovchinnikova\, Andrei Belianko
 u
URL:https://pretalx.com/pyconde-pydata-2026/talk/EWZMJK/
END:VEVENT
END:VCALENDAR