Inspiration
We were given a challenge: do something about burnout in IT workers. But instead of jumping straight to a technical solution, we decided to approach it from a design perspective with humans at the center. We started by asking why so many existing tools fail: they monitor, they report, they alert managers. They treat burnout as a productivity problem rather than a human one.
We wanted to do the opposite. The question we kept asking was: what would this look like if we designed it for the person, not the organization? That led us somewhere more honest: the best intervention for burnout isn't a survey or a manager's dashboard. It's getting people outside together, before the damage is done.
What we built
UNPLUG is a Slack bot with six components we call organs. The Radar collects team-level behavioral signals every 5 minutes: message timestamps, after-hours volume, channel fragmentation, and social health (days since the last team hangout). The Brain runs an Exhaustion Index formula that normalizes against each team's own 14-day baseline, so it adapts to how your team actually works. When the index crosses the team's threshold, the Voice uses Gemini 2.0 Flash to generate a weather-aware, location-specific activity suggestion. The Heart handles voting: if 3+ teammates react with 👍, it's happening. The Garden grows with every shared experience, turning photos from hangouts into collective memories. The Mirror is the privacy layer: blind tokens, zero management access, and a /unplug-forget command that deletes everything.
We also built a live visual dashboard that shows the garden blooming or wilting in real time based on team health signals from Slack.
How we built it
Node.js with the Slack Bolt SDK for the bot, Gemini 2.0 Flash for activity generation, Open-Meteo for live weather (free, no API key needed), and node-cron for scheduling. We used Claude, Gemini etc.
Challenges
The hardest design challenge was the surveillance question. Any system that watches employee behavior can easily become a tool for control rather than care. We solved it architecturally: individual IDs are counted in memory and immediately discarded, only team aggregates hit disk, and management has zero API endpoints. There is no admin panel by design, not by policy.
The second challenge was the garden mechanic: how do you know the team actually went outside? We decided not to verify it. The team votes to go, and if they vote, the garden trusts them. The garden is a collective memory album, not a compliance system.
Built With
- claude
- note.js
- python
Log in or sign up for Devpost to join the conversation.