Towards an In-context LLM-based Approach for Automating the Definition of Model Views
summary
Proposes an in-context LLM pipeline that converts EMF metamodels to PlantUML-like text, feeds three prompt templates to a GPT-4o-class model via LangChain, and parses JSON outputs into VPDL draft viewpoints; requires only the metamodels plus a short natural-language view description, avoids RAG and fine-tuning, and uses few-shot chain-of-thought (1–5 examples). Evaluated on four existing views, produces 100% syntactically correct VPDL drafts with one exact join match and one correct-but-order-swapped join, integrates with EMF Views artifacts and code are available.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:43Model views and EMF
- 2:31Problem with book example
- 5:41LLM automation proposal
- 7:56Solution breakdown
- 12:49Evaluation results
- 16:03Key contributions
- 18:38Future work
- 21:09Q&A
related talks
43mAI Development Lifecycle: Learnings of What Changed with LLMs
46mBeyond Basic Prompting: Supercharging Open Source LLMs with LMQL's Structured Generation
26mLanguage Model Agents Enable Semi-Formal Programming
22mTowards Verified Code Reasoning by LLMs
21mChatGPT, LLMs, and LangChains - A Beginner's Guide
53mSemantics-based program manipulation in the age of LLMs
48mBeyond the Prompt: Evaluating, Testing, and Securing LLM Applications
28mExplaining Machine Learning Models with LLMs