Novel Preprocessing Technique for Data Embedding in Engineering Code Generation Using Large Language Model

We introduce four principal contributions to augment the capabilities of Large Language Models (LLMs) in generating domain-specific code: (i) leveraging LLM-based data splitting and data renovation techniques to refine the semantic representation within the embedding space; (ii) proposing an effective method for refactoring existing scripts, enabling the generation of new and high-quality scripts with the aid of LLMs; (iii) developing the Implicit Knowledge Expansion and Contemplation (IKEC) Prompt technique; and (iv) showcasing the efficacy of our data pre-processing approach through a case study using engineering simulation software RedHawk-SC. Our contributions collectively advance the Retrieval-Augmented Generation (RAG) framework, enabling more relevant and precise information retrieval. An arena-style evaluation by 28 domain experts and 182 votes confirms the significant effectiveness of our methods. Notably, our approach achieves up to 1.43 times the improvement in code generation for MapReduce applications compared to the Chain-of-Thought (CoT) technique.