fedmatch | Devpost

Inspiration

For the first time in history, the Federal government has made it a goal that 40 percent of the overall benefits of certain Federal climate, clean energy, affordable and sustainable housing, and other investments flow to disadvantaged communities that are marginalized by underinvestment and overburdened by pollution. As a result there are billions of dollars out there specifically to serve disadvantaged communities. Although the money is technically there, the sheer amount of grants available, in addition to the technical difficulty of filling out grant applications, has made it that money still doesn’t reach those who need it the most and who can use it most effectively.

What it does

The current solution (Grants.gov) is regularly updated with a plethora of opportunities (2800+ multipage documents right now!), but each grant only targets a comparatively small audience. As such, to understand grant eligibility and fit, users must spend hours parsing lengthy application notices to identify potential fits — time that these entities must pay grant writers for, that could have gone to actual grant-writing. FedMatch aims to address the difficulty of searching through thousands of available grants by leveraging AI in order to improve the discoverability of federal grants that are relevant to your project and that you have a high likelihood of receiving.

How we built it

First we had to build a python web scraper that downloaded documents relevant to every grant available on grants.gov.
Then we loaded all the documents, vectorized them, and stored them in a RAG using Chroma and LangChain.
We use the RAG to give relevant context to the prompt. Additionally we added a template for the AI to follow to produce useful information.
We built our frontend in React
We built our backend on Flask

Challenges we ran into

Submitting on time
Communicating effectively and without conlict
Parsing web data was tedious and had lots of edge cases.

- Parsing through Grants.gov took 4 hours for a python program. Imagine how hard it is for a human to do.

Accomplishments that we're proud of

Creating a script that is able to scrape almost all grant related information off of Grants.gov and ingesting it into a RAG.

What we learned

We learned how to use premier AI tools from our sponsors.

What's next for fedmatch

Reimagining the UI: We focused so much on data collection and building the RAG, so we didn't have enough time to focus on our front-end.
Ingesting data that's beyond Grants.gov into our RAG. Examples of relevant information that is missing on Grants.gov is: number of applicants who applied, how many grants are available to your state/city, examples of successful proposals.
If we are able to ingest examples of successful proposals, we could work with technical assistants and grant writers to help increase their capacity to help their communities.
One limitation of tools such as FedMatch that aim to democratize information is that they often end up benefitting those who are already advantaged, thus exacerbating existing inequities. It is important to consider ways in which our product might create harm even and take the necessary actions to avoid that. This will require more time to think over than just a weekend.