Guarding LLM-aided Software Transformation Tasks via Component Exoskeletons
Name
3766882.3767171.pdf
Size
576.35 KB
Format
Adobe PDF
Checksum (MD5)
682046a4174f7e5eee6a049ed8102cb9
Author(s)
Lamprou, Evangelos
Kalhauge, Christian
Rinard, Martin
Vasilakis, Nikos
Date Issued
October 13, 2025
Publisher
ACM|Practical Adoption Challenges of ML for Systems
Citation
Evangelos Lamprou, Christian Gram Kalhauge, Martin C. Rinard, and Nikos Vasilakis. 2025. Guarding LLM-aided Software Transformation Tasks via Component Exoskeletons. In Proceedings of the 4th Workshop on Practical Adoption Challenges of ML for Systems (PACMI '25). Association for Computing Machinery, New York, NY, USA, 13–18.
Version
Final published version
Abstract
Large language models (LLMs) are achieving state-of-the-art results across a wide variety of software transformation tasks---including translating across languages and lifting opaque software components to high-level languages. Unfortunately, their results are often subtly incorrect, insecure, or underperformant---affecting the widespread deployment of these LLM-driven techniques in settings that go beyond the narrow scope of academic papers. This paper posits that such widespread deployment crucially depends on developing appropriate model guardrails for safeguarding the results of the transformation process. Such guardrails can be supported by component exoskeletons, tunable partial specifications extracted mostly automatically from the original, pre-transformed component. Exoskeletons serve as component projections that supplement, and often go through, the entire transformation process, confirming that the new, transformed component meets the original specifications. They show promise on several real-world scenarios and unearth exciting research directions.
Description
PACMI ’25, October 13-16, 2025, Seoul, Republic of Korea
MIT Department
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1145/3766882.3767171
