Inspiration

GrantMate was born after hearing PHC’s grant writers describe the difficulties of tailoring answers for each sponsor while double-checking every data point manually. We wanted a copilot that could speak in PHC’s voice, honor the organization’s programs, and keep responses grounded in truth—no hallucinations, no generic fluff.

What it does

The app turns pasted grant questions into tailored draft responses sourced directly from PHC’s knowledge base. It summarizes sponsor context, scores the fit out of 5, explains the rationale, and cites the exact documents used so writers can refine and submit with confidence.

How we built it

We built a FastAPI backend wrapped around a retrieval-augmented generation pipeline that prioritizes PHC’s quantitative and qualitative archives. Gemini handles drafting while guard rails validate outputs. A React + Vite front end renders dual panels for questions/answers, a dynamic fit summary, and one-click export-ready copy.

Challenges we ran into

CORS misconfigurations, dependency drift, and evolving API schemas were early headaches. The bigger lift was designing guard rails to block hallucinations and ensuring sponsor-specific context never invented programs PHC doesn’t run. We also tuned the UX so low fit scores read as useful guidance, not failure.

Accomplishments that we’re proud of

Every generated answer stays anchored to PHC’s real work, complete with citations and an honest fit narrative. We deliver human-ready drafts in seconds, calculate alignment scores without sugarcoating, and explain how responses were tailored—turning weeks of manual rewriting into a focused review session.

What we learned

GPT-sized models shine when grounded in curated data and wrapped with transparent explainability. By investing in RAG, validation loops, and consistent UI feedback, we proved AI can amplify—but not distort—how PHC presents itself to funders. Trust is earned through truthful outputs and clear design.

What’s next for GrantMate

We’re adding richer analytics for grant prioritization, enabling collaborative editing, and expanding the knowledge base to cover new programs in real time. The long-term vision is a full fundraising control tower where tailored narratives, fit scoring, and real-world proof live side by side.

Built With

Share this project:

Updates