A Dual-Track Framework for Template-Constrained LaTeX Conversion

The article presents a Dual-Track Framework for converting Markdown drafts into LaTeX, addressing limitations of existing deterministic and end-to-end LLM approaches. This framework separates template formatting from document processing, utilizing an offline track for template constraints and an online hybrid pipeline that combines LLMs for reasoning tasks with rule-based engines for deterministic processing. Empirical evaluations show that this method improves structural fidelity and compilation success rates across multiple LaTeX templates, making it a significant advancement for practitioners in document conversion tasks.