Software, precisely built.
Artist Opportunity Classification With AI
Culturalyst is digital infrastructure for both discovering local culture and sharing the resources that can help it thrive.
Since our last engagement, the network has grown to 12 sites encompassing most of Louisiana, Connecticut, and even a site in Albania! And with that growth, a lot of lessons have come.
Pain Point
While the Culturalyst platform allows organizations to join and post artist opportunities, such as an open call for sculptors or painters, the arts councils are generally too understaffed to find and input them into the system. But there are opportunities scattered all over the web. Sam Bowler, CEO at Culturalyst, wondered whether those opportunities could be crawled, like Google does for search.
Opportunities - Imported by Culturalyst (left) - Input by an Arts Council (right)
Showstopper
We discussed it at length, and spent some time with tools that could fetch those opportunities once they were identified. The whole idea ground to a halt when we realized that we were just moving the staffing problem to Culturalyst. We could get the opportunities posted on random arts and culture sites, but a human would need to classify and translate them into our domain model and taxonomy.
The classification and transformation steps are crucial, since an artist who subscribes to an email feed of opportunities in their region, medium, and genre expects relevant material. A jazz saxophonist has a very specific requirement and doesn't want to see opportunities for sculptors unless they are also a sculptor and have selected it in their profile. Some opportunities may be globally open, some locally focused, so regional classification is an important feature for this network for local culture.
Barrier Removed
The project sat on a back burner until the advent of OpenAI's GPT-4 API. We thought that the showstopper that shelved this project might now be surmounted.
In the end, we built a classification and transformation system using OpenAI API / SDK, JavaScript, and Node.js.that uses few-shot learning with examples of output along with our entire taxonomy, and opportunity data model for reference. The result could take opportunity data of arbitrary shape and create valid and reasonably translated database entities.
This means more relevant opportunities for artists in the Culturalyst network.
Culturalyst
- View Opportunities on the New Orleans Site.
- Read about our previous work with Culturalyst.