A tangible AI oracle built from a Rubik's Cube.

When you're stuck on a problem, the worst thing you can do is keep thinking about it the same way.

Most AI interfaces are linear — you type, it responds, you go deeper into the same thread of thought. Unreasonable Cube is an attempt to break that pattern.

You speak a challenge into the cube. It maps your challenge to six perspectives — the different facets of how humans naturally understand a problem: emotional affinity, relationships, timing, systemic forces, blind spots, and the angle you haven't named yet. Each perspective is assigned to a face of the cube.

You then rotate the cube's gears to shift the weight between perspectives, and press a face to select your primary lens. The cube asks up to four follow-up questions shaped by the full weight distribution you've dialed in. Then it speaks a short speculative story — vivid, specific, set seven years from now — that reflects the blend of perspectives you chose.

The experience takes about three minutes. It doesn't give you an answer. It gives you a more honest question.

You speak a challenge ↓ Claude maps it to 6 perspectives (one per face) ↓ You rotate the cube → shifts perspective weights ↓ You press a face → selects primary lens ↓ Claude asks follow-up questions ↓ Claude speaks a speculative future story 

• 5× ESP32 microcontrollers
• 5× 4" ST7796 TFT displays
• 5× EC11 rotary encoders
• Custom 3D printed mechanical enclosure with gear-based encoder decoupling system
• Connected to laptop via USB

Each face has one screen, one encoder, and one ESP32. The gear system lets you rotate the encoder by turning the outer gear ring without touching or stressing the screen behind it.

cubebot_with_hardware/ ├── main.py — main interaction loop ├── config.py — serial ports, model settings, tuning ├── ai_brain.py — Claude API calls (perspectives + story) ├── state.py — conversation state ├── weights.py — Rubik's-style weight rotation logic ├── voice.py — speech input/output ├── serial_comm.py — USB communication with ESP32s └── esp32/ └── cubebot_esp32.ino — ESP32 firmware 

pip install openai-whisper sounddevice numpy anthropic pyserial

export ANTHROPIC_API_KEY=sk-ant-...

Plug in your ESP32s, find their ports:

ls /dev/cu.*

Update SERIAL_PORTS in config.py to map each port to the correct face:

SERIAL_PORTS = { 0: "/dev/cu.usbserial-0001", # red 1: "/dev/cu.usbserial-0002", # orange ... }

python main.py

Built at the MIT Hardware + AI Hackathon

| Tim Qi (Product & Mechanical cube) | Haoyang Li (Mechanical cube & hardware) | Pranav Chaparala (Hardware) | Shu Ou (Software & AI)

Read the full project writeup →