Now What
Now What
Now What
Now What? is an activity generator for parents of young children. Built entirely with AI tooling, it uses location, weather, safety alerts, and child age to deliver a single, developmentally appropriate suggestion.
Now What? is an activity generator for parents of young children. Built entirely with AI tooling, it uses location, weather, safety alerts, and child age to deliver a single, developmentally appropriate suggestion.
Category
Education
Category
Education
My role
Product Design
My role
Product Design
Timeline
2 weeks
Timeline
2 weeks
Problem
Parents have limited time to plan activities.
Parents of young children need activity ideas but rarely have the time or mental bandwidth to sift through Google searches or filter through local event sites.
Solution
A simple activity generator built for decision fatigue.
Now What? eliminates the need for research and adds an element of discovery with a single suggestion. It required safety-constrained prompt design, a secure API architecture, and a graceful fallback system to address an unsolvable hyperlocal event-data gap.
Process
Product decisions
I gave the app an enthusiastic, authentic, and knowledgeable voice by training the model on explicit examples of what to avoid and include. To reduce friction, user settings persist across sessions so people don’t have to re-enter their information each time.
The data problem
The key assumption that reliable, free, hyperlocal data on family events existed proved false. To address this gap, I used a fallback hierarchy: Google Places for real venue references, defaulting to home/backyard activities when live event data isn’t available.
Technical architecture
Now What? pairs a React/Vite/Tailwind frontend with an Express proxy (to secure the API key) and a prompt-engineering layer. It integrates Open-Meteo, NWS, Google Places, and Figma MCP for weather, alerts, venue data, and design-to-code translation.
Problem
Parents have limited time to plan activities.
Parents of young children need activity ideas but rarely have the time or mental bandwidth to sift through Google searches or filter through local event sites.
Solution
A simple activity generator built for decision fatigue.
Now What? eliminates the need for research and adds an element of discovery with a single suggestion. It required safety-constrained prompt design, a secure API architecture, and a graceful fallback system to address an unsolvable hyperlocal event-data gap.
Process
Product decisions
I gave the app an enthusiastic, authentic, and knowledgeable voice by training the model on explicit examples of what to avoid and include. To reduce friction, user settings persist across sessions so people don’t have to re-enter their information each time.
The data problem
The key assumption that reliable, free, hyperlocal data on family events existed proved false. To address this gap, I used a fallback hierarchy: Google Places for real venue references, defaulting to home/backyard activities when live event data isn’t available.
Technical architecture
Now What? pairs a React/Vite/Tailwind frontend with an Express proxy (to secure the API key) and a prompt-engineering layer. It integrates Open-Meteo, NWS, Google Places, and Figma MCP for weather, alerts, venue data, and design-to-code translation.
Main Flow
Main Flow
Prompting
Prompt Engineering & Safety
During testing, the model hallucinated an activity suggestion that wasn't safe for the target age group. I rewrote the safety constraints with more precise language to close that gap, then built a regression-testing protocol to check future prompt changes against it. The process caught the original issue and has held since.
Prompting
Prompt Engineering & Safety
During testing, the model hallucinated an activity suggestion that wasn't safe for the target age group. I rewrote the safety constraints with more precise language to close that gap, then built a regression-testing protocol to check future prompt changes against it. The process caught the original issue and has held since.
Reflections
What I learned
I always learn the most from user feedback. I enjoyed making something useful for real people without knowing how to code. I showed other moms my prototype, got their feedback, and iterated in real time, adding fee transparency and explaining each activity's developmental benefit based on what they told me. That loop moved faster with AI tooling than any iteration timeline I'd worked with before. Shipping proved I could build and respond to real users on my own. I learned that the feedback loop, is just as valuable as code.
Reflections
What I learned
I always learn the most from user feedback. I enjoyed making something useful for real people without knowing how to code. I showed other moms my prototype, got their feedback, and iterated in real time, adding fee transparency and explaining each activity's developmental benefit based on what they told me. That loop moved faster with AI tooling than any iteration timeline I'd worked with before. Shipping proved I could build and respond to real users on my own. I learned that the feedback loop, is just as valuable as code.
