ImpactLink
Multi-agent platform for nonprofit grant discovery, proposal generation, funder-specific revisions, and budgeting.
At a glance
Trojan Hacks — Social category
Passing frontend and backend tests
Stateful proposal workflows with human review
See it in action
What needed solving?
Nonprofits often face a fragmented grant application process: finding relevant funding opportunities, interpreting eligibility requirements, writing detailed proposals, and preparing budgets. These tasks require significant manual research, coordination, and repeated revisions.
What I built
Co-developed the original hackathon prototype, then independently rebuilt and expanded most of the frontend and backend. Implemented grant retrieval and ranking, stateful AI proposal workflows, funder-specific revisions, human review, and structured budget preparation.
Overview
ImpactLink is an AI-powered grant preparation platform that helps nonprofits move from an initial project idea toward a grant application. It combines funding opportunity discovery, guided proposal drafting, funder-specific gap analysis, targeted revisions, and structured budget preparation. Human review is integrated throughout the workflow.
System architecture
Explore the workflow
Use semantic retrieval with PostgreSQL pgvector, AI reranking, and recorded eligibility checks to identify potentially relevant funding opportunities.
Decisions, validation & lessons
Engineering decisions & tradeoffs+
Stateful workflows with human review
Used LangGraph to coordinate multi-step proposal drafting and revision while preserving session state and allowing users to inspect and revise generated content.
Tradeoff: Stateful workflows introduce complexity around checkpoints, session recovery, and keeping user edits synchronized with AI-generated content.
Semantic retrieval with AI reranking
Used embedding-based retrieval with PostgreSQL pgvector to identify relevant grants, followed by AI reranking and eligibility checks to refine the results.
Tradeoff: Semantic relevance does not guarantee eligibility. Missing, outdated, or funder-specific requirements still require verification.
Deterministic budget calculations
Separated AI-assisted interpretation and drafting from financial arithmetic. Python calculates budgets using explicit cost assumptions, making calculations reproducible.
Tradeoff: The application requires structured financial inputs and additional validation rather than relying on free-form AI-generated estimates.
Validation & results+
The repository documents 144 passing backend tests, 63 passing frontend tests, and a successful frontend production build as of November 5, 2026. Earlier retrieval and drafting evaluations used a small set of synthetic scenarios and are not production-quality benchmarks.
Won the Social category at Trojan Hacks. Independently expanded the original hackathon project into a full-stack prototype supporting grant discovery, proposal development, funder-specific revisions, and budget preparation.
Limitations & learnings+
ImpactLink is a prototype, not a submission-ready grant application service. Eligibility checks cover only recorded requirements, AI-generated proposals require human verification, and database-level tenant isolation and broader real-world evaluation remain future improvements.
Explored multi-agent orchestration, retrieval and ranking quality, stateful workflows, and human-in-the-loop design. Learned to separate probabilistic AI tasks from deterministic financial calculations while balancing automation, traceability, and user control.