Phase 1 Pancreatic Cancer Trial: Efficient LLM Document Guidance

A Phase 1 oncology trial advances only when a sequence of large regulatory and clinical documents is authored, reviewed, and approved, while patients wait for that paperwork to be assembled. This paper illustrates a repository based large language model (LLM), driven by a single master prompt that writes and executes a schedule of sub-prompts committed to GitHub in real time; which can hasten the entire Phase 1 process by generating documents through an auditable "mermaid" to "draft" to "full" to "final" pipeline. The method was conducted on a prior pancreatic ductal adenocarcinoma program: an on-premises LLM directed robotic Whipple with perioperative daraxonrasib (RMC-6236), whose Phase 1 and Phase 2 protocols were themselves single-prompt multi-file builds. The following trial aspects were addressed: the initial IND and IRB package; protocol amendments with synchronized consent; cohort-review packages; the complete clinical-hold response; the Phase 2-to-3 briefing package; and the CSR and NDA/BLA modules. Only the administrative and preparation time bucket compresses, while clinical follow-up and fixed regulatory review clocks do not. Papers finish in 1-4 days, based on prompt and input status. Twenty-four colored Mermaid converted figures ground the prose in LaTeX. Paper repository, directory, and file name samples were verified manually by the author. For terminal patients, probable benefit exceeds probable risk.