GrantSniper
AI grant discovery and proposal drafting, productized into a recurring commercial workflow — as the sole technical owner.
DE-IDENTIFIED BY CATEGORY + PROBLEM · NO CLIENT NAME, NO REAL GRANT DATA · RELATIVE METRICS ONLY
Discovery and drafting, as one workflow
A biotech client was losing time to grant discovery and first-draft proposals. The brief: productize both — surface the right calls, then draft credible, citable proposals — as a workflow they'd pay for month after month.
Commercial, recurring, trustworthy
Subscription-grade reliability, drafts a scientist could defend in front of funders, and retrieval grounded in real sources — not confident fiction.
- Grounded retrieval — every claim traceable to a source
- Drafts a human can defend, not just generate
- [PLACEHOLDER: OTHER HARD CONSTRAINTS]
The decisions that mattered
[PLACEHOLDER: 2–3 SHORT PARAGRAPHS ON WHAT YOU DELIBERATELY LEFT OUT — e.g. WHERE YOU KEPT A HUMAN IN THE LOOP INSTEAD OF FULL AUTOMATION, WHAT YOU REFUSED TO AUTO-SUBMIT, THE SCOPE YOU CUT TO SHIP THE RECURRING WORKFLOW.]
The shape of the system
Where it got hard
[PLACEHOLDER: THE MESSIEST SLICE — HALLUCINATED CITATIONS, RETRIEVAL DRIFT, AN EVAL THAT CAUGHT A BAD DRAFT. WHAT YOU MEASURED, HOW YOU DEBUGGED IT, WHAT YOU CHANGED.]
Outcome, in relative terms
[PLACEHOLDER: RELATIVE, DE-IDENTIFIED]
[PLACEHOLDER]
[PLACEHOLDER]
Directed by agents, reviewed by hand. Retrieval and drafting are evaluated, not trusted — with a human on every claim that reaches a funder.