-
-
Figma work
Inspiration
While we were brainstorming, our team initially tried to imagine ways that AI could help improve socioeconomic equity. One of these ideas was using LLM’s to personally manage professional networks the way a high net-worth individual may have an assistant who schedules dinners or meetings with important contacts. As we thought about this application deeper, we realized the implications of using AI to improve relationships outside of the professional world. This tool could be incredibly helpful for those who struggle to initiate conversations due to social anxiety or neurodivergent conditions that affect social literacy. Further research reveals a severe lack in Social Skill Training (SST) for the digital age and tools to help navigate it; a gap which SocialSpark aims to fill.
What it does
Our app allows users to create a list of connections that they’d like to grow a stronger relationship with. Based on the information given about these connections such as descriptions, desired closenesses, and frequencies of contact, SocialSpark generates helpful personalized tips for starting conversations that relate specifically to each connection’s interests as well as maintain researched social boundaries.
How we built it
SocialSpark is a native SwiftUI App which communicates with our FastAPI backend architecture and MySQL database. We used Figma for UX design. Additionally, we used Google Gemini and Perplexity models for real-time suggestions to help guide conversations.
Challenges we ran into
We ran into some initial issues with authorization token verification which took some time to work out. Additionally, the integration between our user interface app, web server, and background database was a lengthy process due to some unfortunate errors. Another issue was dealing with API rate limits and safeguards when building out our conversation suggestion features.
Accomplishments that we're proud of
We're very proud of the Contact Priority Rating System we developed based on psychological research about interaction frequency based on relationship types. Our robust system is able to generate helpful priority scores for contacting connections that the generative models making conversation starter suggestions could use to be more effective.
On that note, our integrations with Google’s Gemini models and Perplexity’s model suite were particularly interesting because we spent a lot of time prompt engineering the model to effectively use our Connection information and current events news to generate concise and accurate suggestions.
What we learned
Our team learned a lot about new development practices as many of the technologies we used were novel to us such as IOS development with Swift, advanced LLM prompt engineering, and UX design with Figma.
What's next for SocialSpark
We aim to improve on and fine tune our models with real user data. As more people use SocialSpark, we can get an idea of the general social needs of our target audience and provide more features.
Log in or sign up for Devpost to join the conversation.