Get Started
Research questionHow can LLMs recover architectural design rationale from implicit source-code commits?Architectural decisions are often embedded indirectly in unstructured commit changes and messages instead of being documented as explicit rationale. This makes it difficult to distinguish the architectural motivation from implementation details.
AI
Natural Language Processing
Technology
Latest papersRecent research connected to this question, newest first.Can LLMs Extract Architectural Design Decisions from Source Code Commits? - A Preliminary Exploratory StudyThe preliminary evidence covers zero-shot and few-shot prompting with four LLMs on 30 developer-written architectural design decisions from open-source projects. Outputs showed strong textual alignment overall, but were often overly long, implementation-focused, and missing the underlying rationale; one author manually reviewed outputs from one model.research paper · Sep 3, 2026
Related questions
How can LLMs edit another model’s code while preserving functionality and minimizing unnecessary changes?How can we detect when LLM-decompiled code diverges from original behavior or erases disclosed vulnerabilities?How can LLM-based automated program repair detect patches and artifacts unsupported by repair evidence?How can industrial root-cause analysis reconcile conflicting free-text hypotheses without slow, deeply agentic LLM workflows?
Home
Topics
Search
Library