This formative evaluation describes a governed, large language model-supported workflow for grant discovery and proposal development at a Colombian health-AI company. The workflow runs in four phases (opportunity discovery, eligibility triage, Win Readiness scoring and proposal drafting) and relies on repository state, issue labels, deduplication rules and researcher review checkpoints to turn broad funding searches into prioritized work items.
The workflow covered 20 funding scenarios, more than 50 search combinations and over 200 reviewed opportunities. Across 10 logged scouting runs it scanned 741 sources, kept 83 candidates after filtering, opened 46 net-new issues and flagged 220 duplicates; the Win Readiness Score was refined as a relative ranking across 49 grants, and five applications reached interview or finalist stage between March and June 2026. The authors frame this as a formative evaluation rather than a causal claim about grant success, and the work is a preprint not yet peer reviewed.
