Generative
AI Party
Planner
A speculative project redesigning Partiful’s Party Genie tool by integrating Generative AI for more customized event creation and expanding the tool to web platforms. Live at dub.sh/partyvibe.
.png)
Key Skills
AI Prompt Design
Wireframing
Animated Prototypes
Roles
Product Designer
Prompt Engineer
Identifying Opportunities
In this speculative project, I partnered with a front end developer to redesign Partiful's Party Genie tool. Partiful is one of the fastest growing event planning platforms today, allowing users to easily make fun invitations for any event. In 2024, the company released a “Party Genie” tool on its app that gave users inspiration for their next party. Users select one category from a list of terms such as “chill” or “summer”. Once they choose a category, an event pops up. If users like the event, they can press a button to generate an e-vite.
A video showcasing how users currently can use the Party Genie tool in the Partiful app. All events are pre-written by a copywriter and each category only features 5 events.
I then generated a list of strengths and weaknesses of the current platform.
Strengths
-
Provides inspiration for events
-
Demonstrates a strong brand voice
-
Makes it easy to generate a full e-vite
Weaknesses
-
Ideas are not personalized to the user’s needs
-
Creates a limited number of events
-
Only available through the app
Through this exercise I was able to narrow my focus to three main approaches:
1. Expanding the Tool for Web Desktop Platforms for Wider Reach
2. Using Large Language Models for More Personalized Event Creation
3. Integrating Generative AI to Facilitate a Larger Number of Options per Selection
Web Design
My first step to designing Party Genie for the web was to understand the company’s design system. Using my browser’s developer tools, I reverse engineered elements of their approach to typography, content structures, layouts, and buttons. Based on this design system and Partiful’s existing web experiences, I developed wireframes and high-fidelity prototypes
I built wireframes for desktop and mobile that ensured a consistent browser experience across devices.
.png)
.png)
I developed high-fidelity prototypes that allowed users to select multiple categories instead of only one.
One of the most important elements of my new Party Genie tool was ensuring that users were able to add their own categories. To make this possible, I created an “add vibe” button that users could click and type out their own category.
Using Figma, I created animated prototypes to demonstrate user flows and interactions including the ability to add categories.
Integrating Generative AI
First, I developed a multi-shot prompt for Google's Large Language Model that could produce unique party ideas based on keywords from the user.
The multi-shot prompt I developed for Google Gemini generated outputs for five categories: Event Title, Description, Event Image, Background Template, and Animation
To ensure that I maintained Partiful’s unique and witty brand voice, I provided the AI model with example responses based on Partiful’s own event inspiration site.
However, user testing revealed a significant issue: requests using the same keywords resulted in nearly identical outputs from the LLM, with only minimal changes to the event title or description
Event #1
Event #2
In our original system, the LLM would generate extremely similar events if the same categories were selected. The two events pictured above show that only small changes to the event titles and descriptions occurred between requests.
To address this, I drew inspiration from Automated Prompt Optimization. Working with the software developer, we implemented a system to automatically refine user prompts. This system adds a randomly chosen noun to each prompt before submitting it to Google Gemini. For example, if a user submitted the keywords “birthday, wine”, the program would actually send to the LLM “birthday, wine, dog”. If the user submitted those same keywords a second time, the program would send a slightly different prompt like “birthday, wine, hopscotch”.
Three examples of AI-Generated events with our updated prompting system. Using automated prompt optimization the program can generate an unlimited number of unique events using the same categories.
Final Thoughts
This project was an exciting opportunity to not only develop my visual design and device conversion skills, but also to dive head first into prompt engineering and applying generative AI in non-traditional settings. For the future, I will be focusing on conducting user research studies such as focus groups, user surveys, and A/B testing for UI features. Furthermore, I want to investigate into using AI image generation tools that can create unique posters for each event rather than rely purely on the ones that Partiful has on their site.
