Artist Grant Hub

Inspiration

I watched my partner go through the grueling process of applying for grants as a small non profit organization, after multiple all nighters and the nuisance of only successfully receiving less than 10% of the grants they applied. It was heart wrenching. And then I found out that things are even more dismal for individual artists especially neurodivergent creators who find the application process an overwhelming source of anxiety and shame. The goal for this tool is to create a framework that supports creative integrity and access.

What it does

The tool has 3 agents, with 4 tasks.

A researcher agent, a copy writer agent, and a project management agent. The researchers core function is to analyze raw, unstructured data related to grant opportunities often aggregated through scraping or data extraction and transform it into actionable kanban boards, filtered, and precisely structured insights for a specific end-user, such as an emerging artist or a small nonprofit. The copy writer is tasked with generating compliant, persuasive, and clear drafts for grant components (Project Description, Background, Outcomes) using unstructured input and the grants qualification details. And the project manager transforms structured grant data into actionable micro-project steps, key dates, and a visual task management format (like a Kanban board or timeline), ready for a calendar system.

The researcher agent has 2 tasks associated to it, the first grant discovery and then the 2nd is grant evaluation.

How we built it

It was built using lighting.ai, I found an crewai environment that leveraged serperdevtool as it's search tool. I added the roles and tasks for each agent along with the sequence.

Challenges we ran into

I had trouble adding a calendar tool to the environment, and I had trouble aligning the agent sequence with my UX flow for a user. For instance at one point, the agentic sequence would return all search results and provided copy and project plan for all of them instead of just the users selection. I added additional context, and human input to one of the tasks to act as pause to check for use input, but that had to be manually typed.

Accomplishments that we're proud of

I really enjoyed the process of thinking through the sequence steps for each agent/tasks.

What we learned

It's generally better to keep tasks as one action instead of bundling them. At one point I had my research agent doing both discovery and evaluation in a single task it, I was getting constant errors until I split them.

What's next for Artist Grant Hub

Continue building out the agentic solution and develop a UX for user input as a v0.

Built With

• lighting.ai
• python