BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//132.216.98.100//NONSGML kigkonsult.se iCalcreator 2.20.4//
BEGIN:VEVENT
UID:20260522T221648EDT-5705ROZeNn@132.216.98.100
DTSTAMP:20260523T021648Z
DESCRIPTION:Abstract\n\nThe rapid advancement of large language models (LLM
 s) has sparked their application across diverse domains\, including softwa
 re engineering (SE). In SE\, LLMs have shown strong performance on a range
  of tasks and enabled the development of practical assistance tools such a
 s GitHub Copilot and Cursor. However\, integrating LLMs into complex real-
 world applications remains a significant challenge\, particularly for doma
 in-specific tasks that are underrepresented in LLM training data.\n\nHuman
  domain experts typically follow systematic domain-specific best practices
  when approaching complex tasks\, relying on comprehensive workflows and a
  variety of domain-specific tools. To enhance LLMs’ ability to handle such
  tasks\, existing research has focused on integrating these expert practic
 es into LLM-based applications. Practical frameworks like LangChain and La
 ngGraph have been proposed to support this integration by enabling the con
 struction of domain-specific workflows. However\, the workflow representat
 ions in these frameworks remain limited in their expressiveness\, often la
 cking modularity and the capacity to model complex behaviors.\n\nFurthermo
 re\, domain-specific tools often impose strict constraints on input data. 
 To integrate LLMs with such tools effectively\, it is essential to ensure 
 that LLM-generated outputs conform to these constraints. However\, the inh
 erent nondeterminism of LLMs poses a significant challenge in achieving co
 nsistent outputs with respect to the constraints. While existing research 
 has explored techniques such as constrained decoding to improve output con
 sistency\, these methods primarily target simple output formats and do not
  extend to more complex structures like graphs. Additionally\, the relatio
 nship between consistency and the overall quality of LLM-generated outputs
 \, particularly in graphs\, remains insufficiently understood.\n\nIn this 
 thesis\, I propose two systematic approaches to address the challenges of 
 workflow representation and output consistency in LLM-based applications. 
 The contributions are organized around two high-level research questions. 
 To tackle the first challenge on workflow representation (HRQ1)\, I introd
 uce SHERPA\, a framework that models domain-specific workflows as state ma
 chines\, facilitating the integration of LLMs with domain-specific tools. 
 This framework decouples workflow representation from its concrete impleme
 ntation\, supporting a modular and flexible design of LLM-based applicatio
 ns. Systematic evaluation demonstrates that SHERPA enables rapid experimen
 tation with diverse workflows\, leading to improved task performance and a
  better balance between cost and effectiveness.\n\nTo address the second c
 hallenge on output consistency (HRQ2)\, I propose AbsCon\, a framework des
 igned to ensure the consistency of LLM-generated graphs by leveraging the 
 nondeterministic nature of LLMs. Generalizing a constraint optimization-ba
 sed approach that I originally proposed for scene graph generation\, AbsCo
 n guarantees that the generated graphs satisfy domain-specific constraints
 . Evaluation results further demonstrate that enforcing such consistency a
 lso significantly improves the overall quality of the generated graphs whe
 n compared to human-constructed ground truths.\n
DTSTART:20251024T170000Z
DTEND:20251024T190000Z
LOCATION:Room 603\, McConnell Engineering Building\, CA\, QC\, Montreal\, H
 3A 0E9\, 3480 rue University
SUMMARY:PhD defence of Boqi (Percy) Chen – Domain-Driven and Consistent Int
 egration of Large Language Models: An Input-Output Perspective
URL:/ece/channels/event/phd-defence-boqi-percy-chen-do
 main-driven-and-consistent-integration-large-language-models-input-368031
END:VEVENT
END:VCALENDAR