BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//pretalx//pretalx.devconf.info//devconf-us-2026//speaker//TY3AAX
BEGIN:VTIMEZONE
TZID:EST
BEGIN:STANDARD
DTSTART:20001029T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10;UNTIL=20061029T070000Z
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:STANDARD
DTSTART:20071104T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=11
TZNAME:EST
TZOFFSETFROM:-0400
TZOFFSETTO:-0500
END:STANDARD
BEGIN:DAYLIGHT
DTSTART:20000402T030000
RRULE:FREQ=YEARLY;BYDAY=1SU;BYMONTH=4;UNTIL=20060402T080000Z
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
BEGIN:DAYLIGHT
DTSTART:20070311T030000
RRULE:FREQ=YEARLY;BYDAY=2SU;BYMONTH=3
TZNAME:EDT
TZOFFSETFROM:-0500
TZOFFSETTO:-0400
END:DAYLIGHT
END:VTIMEZONE
BEGIN:VEVENT
UID:pretalx-devconf-us-2026-V8WPUW@pretalx.devconf.info
DTSTART;TZID=EST:20260924T104000
DTEND;TZID=EST:20260924T111500
DESCRIPTION:Performance engineers at Red Hat analyze vLLM inference benchma
 rks across Nvidia\, AMD and TPU accelerators every release cycle. The data
  lives everywhere: benchmark CSVs\, GPU metrics in Grafana\, PyTorch profi
 ler traces\, vLLM source code on GitHub\, and server logs with engine conf
 iguration details. Answering "why did throughput regress by 12% between vL
 LM-0.16.0 and vLLM-0.18.0 for model X?" used to mean hours of manual cross
 -referencing. We built an AI agent to unify it all and do it in minutes.\n
 \nThe AI Performance Agent is a LangGraph ReAct agent powered by Google Ge
 mini\, connected to a FastMCP server exposing 30 specialized tools for ben
 chmark querying\, kernel-level profiler analysis\, source code diffing\, G
 PU metrics\, energy computation\, cost analysis and more\, all through nat
 ural language.\n\nThis talk covers the MCP architecture that makes the too
 ls reusable across any client\, the prompt engineering that teaches the mo
 del when to use which tool and when to stop and ask for clarification \, t
 he Langfuse v3 observability stack\, and deploying the full system on Open
 Shift. Live demo included. \n\nWhat attendees will take away:\n\n- A reusa
 ble architecture pattern for building MCP-based AI agents that integrate w
 ith enterprise data sources\n- Practical prompt engineering patterns for m
 ulti-tool agents that need to be rigorous\, not just fluent\n- A deploymen
 t blueprint for running LangGraph agents with full observability on OpenSh
 ift\n- The open-source tool stack: LangGraph\, FastMCP\, Langfuse\, Stream
 lit\, all community projects
DTSTAMP:20260727T165044Z
LOCATION:Ladd Room (Capacity 170)
SUMMARY:1 Agent\, 30 Tools : What We Learned Building an MCP-Powered Perfor
 mance Detective - Harshith Umesh
URL:https://pretalx.devconf.info/devconf-us-2026/talk/V8WPUW/
END:VEVENT
END:VCALENDAR