Inspiration

Every warning matters because behind every server is someone depending on it.

Uptime Institute’s 2026 outage analysis reports that 57% of surveyed respondents said their most recent major outage cost more than $100,000. But the consequences extend beyond financial losses. Source: Uptime Institute

On July 19, 2022, extreme heat caused both data centers at Guy’s and St Thomas’ hospitals in London to fail. Procedures and appointments were postponed, affecting patients and their families. Source: Guy’s and St Thomas’ NHS Foundation Trust

We created UNICORN to make infrastructure health easier to understand and critical warnings harder to miss. Our goal is to help the people keeping essential services running recognize problems early, understand what is happening, and respond with confidence.

What it does

UNICORN brings environmental monitoring, incident investigation, and a visual facility map into one approachable interface.

Users can explore an interactive 3D data center, select a room to inspect its condition, and identify affected areas through floating red emergency pointers. The dashboard presents temperature, humidity, vibration, cooling status, and recent activity, while incident pages explain detection, response, and recovery.

The backend evaluates incoming readings against safety thresholds, opens incidents, and queues hardware commands. It distinguishes between sending a command, receiving hardware confirmation, and verifying that conditions have recovered.

Our prototype also includes simulated facilities, allowing us to demonstrate overheating, water, vibration, and connectivity problems without putting physical equipment at risk.

How we built it

Hardware

Our hardware design centers on sensing environmental changes and demonstrating a physical response:

  • Raspberry Pi Pico 2 WH: controller for the hardware prototype.
  • Temperature, humidity, and vibration sensors: inputs for detecting abnormal conditions.
  • Cooling fan and PN2222 NPN transistor: cooling response and fan switching.
  • 3-6V DC Gear Motor Set with Fan Blade: demonstration of an adjustable ventilation mechanism.
  • LEDs and buzzer: visible and audible status indicators.
  • SSD1306 0.96-inch OLED display: local status information.

Software

  • Next.js, React, and TypeScript power the application interface.
  • Tailwind CSS and custom CSS provide the pastel visual system and responsive layouts.
  • Blender supplies the campus and unicorn models.
  • Three.js renders the interactive facility, camera controls, room markers, and animated team characters.
  • Python and FastAPI handle telemetry, safety rules, incidents, hardware commands, and simulation.
  • Scikit-learn’s Isolation Forest provides an additional anomaly-detection layer.
  • SpacetimeDB integration supports optional persistence of operational records.
  • REST endpoints and WebSockets provide the backend interfaces for commands and live updates.

Use of GeminiAPI

We implemented an optional Gemini-powered incident investigation agent in the backend.

Through tool calling, Gemini can inspect sensor history, device connectivity, active incidents, and facility status to produce an evidence-based explanation and recommend a response.

The agent can request approved actions such as activating cooling, opening ventilation, or sounding an alarm. These requests remain subject to backend safety checks, and autonomous actions are disabled by default.

Deterministic safety rules remain in control. Gemini cannot bypass them, and the system continues operating when the API is unavailable.

Challenges we ran into

  • Wiring and connecting the hardware components correctly.
  • Getting the cooling fan to activate reliably.

Accomplishments that we're proud of

  • Bringing all the hardware components together into a cohesive final build. What we learned
  • Hardware: How to wire components and integrate sensors and actuators.
  • SpacetimeDB: How to organize and persist telemetry and system state.
Share this project:

Updates

Submission history