BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260624T171342Z
LOCATION:Bldg. 8 - B 102
DTSTART;TZID=Europe/Stockholm:20260701T143000
DTEND;TZID=Europe/Stockholm:20260701T150000
UID:submissions.pasc-conference.org_PASC26_sess165_msa224@linklings.com
SUMMARY:Flowgentic: An Execution Coordination Layer for Adaptive, Agent-Dr
 iven HPC Workflows
DESCRIPTION:Matteo Turilli (IE University)\n\nModern scientific workflows 
 increasingly combine large-scale HPC simulations with machine learning inf
 erence, adaptive decision-making, and dynamic control flow. While agentic 
 orchestration frameworks like LangGraph provide expressive abstractions fo
 r goal-directed coordination, they assume low-latency, interactive executi
 on environments fundamentally incompatible with HPC systems designed aroun
 d explicit task descriptions, static resource requests, and scheduler-medi
 ated execution boundaries. This impedance mismatch forces agentic workflow
 s into static approximations or opaque, monolithic actions, undermining bo
 th their expressiveness and scientific reproducibility.\n\nFlowgentic is a
 n execution coordination layer that mediates between agentic orchestration
  frameworks and heterogeneous HPC execution management systems. Flowgentic
  decouples adaptive decision-making from task execution by externalizing e
 xecution coordination as a first-class system concern. Its architecture is
  organized around four placeable abstractions (Agent, Reasoner, Tool, and 
 Memory) that make the structure of adaptive workflows explicit and observa
 ble without entangling control logic with execution mechanics. Runnable wo
 rk is materialized as executable task ensembles and delegated to downstrea
 m systems like AsyncFlow or Kubernetes, enabling both structured and unstr
 uctured execution backends without requiring backend-specific coordination
  logic.\n\nWe present the design principles, architecture, and initial per
 formance characterization of Flowgentic. We discuss how Flowgentic's co-de
 sign approach supports the broader goal of making AI-enhanced scientific w
 orkflows reproducible, observable, and scalable across facility boundaries
 .\n\nDomain: Engineering, Physics, Computational Methods and Applied Mathe
 matics\n\nSession Chair: Ozgur Ozan Kilic (Brookhaven National Laboratory)
 \n\n
END:VEVENT
END:VCALENDAR