GrantSync

Inspiration

Non-profits often miss out on vital funding simply because they can't effectively navigate the landscape. We realized that while there are over 50 grants available on portals like OurSG, the process of finding the right one is still manually intensive. A non-profit team often spends lots of time writing a proposal, only to get rejected because they missed a single semantic nuance in the funder's mission statement.

We built GrantSync to democratize access to funding strategy. Instead of just a search engine, it acts as an intelligent grant strategist, actively guiding non-profits to align their genuine impact with funder goals and decodes complex requirements into winning proposals.

What it does

GrantSync is an intelligent orchestration platform that helps non-profits find, evaluate, and win funding.

AI Usage Matching: Instead of rigid keyword searches, we use a 2-stage AI pipeline to match your project's core intent against the grant database. This finds opportunities based on shared goals and outcomes (e.g., matching a "Wellness" project to a "Health" grant), ensuring deep alignment even when terminology differs.
The "Grant Coach": Our proprietary Gap Analysis engine doesn't just say "Low Match." It tells you why. It identifies specifically which KPIs you are missing and suggests concrete improvements to your proposal.
Automated Workflow: From Vercel Cron jobs that scrape new grants weekly to automated deadline email alerts (14/7/3/5/1 days out), we automate the administrative burden so teams can focus on impact.

How we built it

Core Stack: Built on Next.js 16 for a responsive frontend and Better Auth for secure session management.
AI Pipeline: We implemented a hybrid RAG (Retrieval-Augmented Generation) system:
- Stage 1 (Retrieval): We use pgvector and semantic embeddings to instantly filter thousands of grants down to the top 20 relevant candidates.
- Stage 2 (Reasoning): We feed those candidates into Google Gemini 3.0 Flash, which acts as a reasoning engine to score them based on Mission, Eligibility, and Impact alignment.
Infrastructure: The app creates a "living" database using Vercel Cron Jobs to scrape government portals, storing data in Neon Postgres via Prisma. Email notifications are dispatched via Resend.

Challenges we ran into

The biggest technical hurdle was the "Embedding Trap." Initially, we relied solely on vector embeddings for matching. We found that embeddings often gave high scores to grants that sounded similar but were actually irrelevant contextually (e.g., a "Research Grant" vs. a "Community Grant" for the same topic).

Embeddings work well for similarity, but not for nuances in the content. We solved this by building a 2-Stage Pipeline: using embeddings for fast retrieval, but adding an LLM (Gemini) logical layer on top to strictly evaluate "Eligibility Checks" that embeddings missed.

Accomplishments that we're proud of

"Glass Box" AI: We resisted the urge to build a "black box." Every AI score comes with a transparent breakdown (Mission/Eligibility/Impact), building trust with non-technical users.
The "Grant Coach": Successfully building the Gap Analysis feature that gives actionable advice (e.g., "Add a metric for senior engagement") rather than just generic feedback.
End-to-End Automation: We built a complete product. From scraping raw data, to AI processing, to Email alerts, and to Kanban tracking.

What we learned

Context is King: Standard RAG is not enough for complex domains like grant funding. You need structured reasoning (LLMs) on top of unstructured search (embeddings) to get usable results.
Agile Evolution: We started with a simple Kanban board and showing Grants available. By iteratively testing with the problem statement, we evolved it into a "Project-First" system where you define your need first, and the system brings the solution to you.

What's next for GrantSync

Post-Award Compliance: Tracking reporting deadlines and impact metric submissions after the money is in the bank.
Auto-Drafting: Using the Gap Analysis data to automatically generate the first draft of the grant application proposal.
Collaborative Writing: Allowing multiple stakeholders to edit the grant proposal in real-time on the platform.