-As someone in the public sector, I've always found it really difficult to find grants relevant for the projects I'm working on. I thought this would be a great opportunity to try to solve that problem while experimenting with SotA models.
-Matches billions of dollars in federal grants to your project and gives you a verdict on how to fit your project to grant requirements.
-Embedded thousands of grant descriptions using medium embed model and saved those embeddings. Used HuggingFace. -Embed proposed project descriptions on the fly. Used HuggingFace. -Use semantic search to score and rank pairings. -Evaluate high ranking pairings with gen AI to give a verdict on whether to apply or not. Used Replicate.
-Getting data out of grants.gov. No API available, just bulk database downloads. -Improving memory usage in Streamlit, solved through use of cache functions in Streamlit and active forums.
-Public embedded search for federal grants for the first time ever. -Using gen AI as an assistant to the prospective applicant.
-Open source AI ecosystem is awesome. Lots of high quality options. -Streamlit continues to be super fun to use.
I'm planning on iterating on Grants Match by improving on the use of gen AI to better match projects and grants based on eligibility. I also want to continue to iterate on the interface to make the process more "delightful".