A Workflow for Grant Discovery and Proposal Development Using Large Language Models: Development and Formative Evaluation

Abstract Background Health-AI organizations in low- and middle-income countries need non-dilutive funding to generate evidence and validate tools in real-world health systems. Grant discovery and proposal development require repeated funder searches, eligibility interpretation, and narrative adaptation. Large language models can support parts of this work, but published guidance emphasizes caution, fact-checking, and human ownership of grant strategy. We did not identify a published description of a governed, repository-based grant development workflow inside a health-AI company. Objective To describe a governed, large language model-supported grant development workflow at a Colombian health-AI company, designed to improve throughput and traceability of grant discovery, triage, and proposal drafting while maintaining researcher oversight. Methods We describe the workflow architecture using internal repository artifacts, issue records, scouting logs, and scoring documentation. The workflow includes four phases: opportunity discovery, eligibility triage, Win Readiness scoring, and proposal drafting. The system uses repository state, issue labels, deduplication rules, and researcher review checkpoints to convert broad funding searches into prioritized work items. Results The workflow covered 20 funding scenarios across foundations, government agencies, pharmaceutical programs, innovation prizes, and bilateral mechanisms. Over 50 search combinations and 200 opportunities were reviewed. A structured scouting log covering 10 runs (2026-05-29 to 2026-06-05) recorded 741 scanned sources, 83 post-filter candidates, 46 net-new issues, and 220 duplicates. The Win Readiness Score was refined as a relative ranking tool across 49 grants. Five applications advanced to interview or finalist stage during the four-month operating period (March to June 2026). Conclusions A governed large language model-supported workflow can make grant discovery and proposal preparation more systematic and auditable for health-AI teams in resource-constrained settings. The evidence supports a formative evaluation, not a causal claim about grant success. Future evaluation should measure proposal quality, researcher effort, and funding secured.