Secure Code Completion Models Tuned for Compliance-Heavy Domains

Abstract

The integration of large language models (LLMs) into software engineering workflows has accelerated code generation and review processes. However, their deployment in compliance-intensive sectors—such as finance, healthcare, and defense—introduces stringent challenges around regulatory adherence,security assurance, and legal accountability. This paper presents a principled methodology for fine-tuning LLMs on domain-specific, security-vetted datasets to produce code aligned with rigorous compliance frameworks. The proposed pipeline addresses dataset curation, model adaptation, and multi-layered evaluation, ensuring both syntactic correctness and regulatory fidelity. Empirical results demonstrate that the fine-tuned models significantly outperform general-purpose LLMs in generating secure, regulation-compliant code.