BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//cfp.devcon.org//devcon8//speaker//W7ZTBE
BEGIN:VTIMEZONE
TZID:IST
BEGIN:STANDARD
DTSTART:20000101T000000
RRULE:FREQ=YEARLY;BYMONTH=1
TZNAME:IST
TZOFFSETFROM:+0530
TZOFFSETTO:+0530
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devcon8-RBDERC@cfp.devcon.org
DTSTART;TZID=IST:20261105T160500
DTEND;TZID=IST:20261105T161000
DESCRIPTION:Platform teams ship LLM features fast. Then production asks a b
 asic question: where did this request go? The answer is rarely one Pod. A 
 chat completion crosses DNS or a Service\, often an ingress or gateway\, t
 hen a model proxy that routes by model name. The serving controller can sc
 ale a Pod from zero. The model server loads weights\, runs inference\, and
  streams tokens back. Each hop has its own failure mode. A 404 on the wron
 g API path. A hang while a cold model wakes. A gateway timeout that kills 
 a long stream. Request-count rate limits that ignore token cost. Teams deb
 ug the symptom because they never mapped the path.\n\nThis talk draws that
  map. We start from a client call to an OpenAI-compatible endpoint. We fol
 low the request into Kubernetes networking\, into the serving proxy\, into
  the Model resource and the Pod that runs Ollama on CPU or vLLM on GPU. We
  use OpenEverest as a concrete case: an Instance CR declares intent\, a pr
 ovider creates a KubeAI Model\, and status writes the phase and endpoint b
 ack. The point is the path\, not the brand. The same hops appear with KSer
 ve or a hand-rolled stack.
DTSTAMP:20261101T132301Z
LOCATION:Community Led Stage
SUMMARY:What Actually Happens When You Send an LLM Request to Kubernetes? -
  Atharva Mhaske
URL:https://cfp.devcon.org/devcon8/talk/RBDERC/
END:VEVENT
END:VCALENDAR