Powerfly
A mobile energy assistant that monitors electricity prices, weather, and household context to recommend when to use energy. I designed and built the frontend, including how the agent explains why its recommendations change.
Honorable Mention · 75 teams
The data existed. Acting on it required constant attention.
Electricity prices change throughout the day, and utilities already publish much of the information people need to make better decisions about when they use energy. The problem is that taking advantage of those changes means actively checking prices, weather, and usage throughout the day. We explored whether an autonomous agent could do that monitoring in the background and surface a recommendation only when there was something useful to act on.
Powerfly used live CAISO electricity pricing and Open-Meteo weather data to generate recommendations as conditions changed.
Kiran and Joseph focused on the backend and autonomous agent loop. Chloe worked on product framing and the pitch. I owned the UI/UX and built the frontend in Next.js and Tailwind. My main design question was how much of a continuously running agent's reasoning a person actually needed to see.
The backend continuously moved through an Observe → Reflect → Plan → Act loop, using electricity pricing, weather, and household context to decide whether there was something useful for the homeowner to do. As conditions changed, it could reconsider and update its recommendation. That created an interface challenge: the user needed to understand what Powerfly recommended now and enough context to understand why that recommendation might have changed.
The Trust Ledger
The center of the design. Each recommendation shows the conditions behind it, so a person can see why a suggestion was made and why it changed. I designed how that history reads at a glance.
Making the loop legible
The agent evaluates conditions continuously. Showing every change would overwhelm the interface, and showing nothing would hide its reasoning. I designed what it should surface and what it should keep in the background.
A single recommendation
Energy data usually arrives as a dashboard. I designed a mobile-first interface that turns pricing and weather into one clear recommendation rather than another screen to monitor.
Designed a history of the agent's recommendations so users could understand why a suggestion was made and what changed when the recommendation updated.
Worked through how recommendations should appear, update, and change as the agent continuously observed new conditions.
Turned pricing, weather, and energy data into a simple recommendation instead of another dashboard someone had to monitor.
Built the mobile-first interface in Next.js and Tailwind and connected it to REST APIs being developed in parallel.
Kiran and Joseph focused on the backend and autonomous agent loop. Chloe worked on product framing and the pitch.
5.5 hours from idea to presentation.
The hour splits inside the 5.5 are approximate.
Working in parallel: because the frontend and backend were being built at the same time, I designed around the expected agent response first and adjusted the interface as the actual data structure took shape.
What we ran into, and what I'd change.
We narrowed to the strongest data source
We initially explored electricity, water, and gas. Reliable real-time data was much easier to access for electricity, so we narrowed the MVP instead of simulating data for the other utilities.
Surfacing every change would overwhelm the interface
The agent could continuously evaluate new information, but surfacing every change would quickly make the interface difficult to scan.
The agent's output changed during parallel development
The backend and frontend were being developed in parallel, so the structure of the agent's output changed as the system evolved.
An energy assistant that monitors conditions in the background.
Recommendations
Powerfly surfaces a recommendation when changing conditions create an opportunity to use energy at a better time.
Trust Ledger
Each recommendation includes the conditions behind it, making it possible to understand why the agent made or changed a suggestion.
Updates
When conditions change, the recommendation can change with them and the interface shows what influenced the update.
Background monitoring
The agent continues observing electricity prices, weather, and household context without requiring the homeowner to monitor a dashboard.
Powerfly was my first hackathon where I focused entirely on UI/UX and frontend while other teammates built the system behind it. Designing for a continuously running agent was different from designing a traditional dashboard. The challenge was not displaying everything the system knew. It was deciding which decisions were worth surfacing, what context someone needed to understand them, and what could stay in the background. It made me think about transparency as an interface decision, not simply showing more information.
Four people, 5.5 hours from idea to prototype.
Michelle Bui led UI/UX and the frontend. Kiran and Joseph built the backend and agent system. Chloe led product framing and the pitch.