MurMur

• 

  hardware

• 

  hardware2

Inspiration

AI is becoming more capable every day, but interacting with it still requires pulling out a device, opening an app, and typing or speaking. We wanted to explore a faster, silent way to interact with AI, using subtle jaw movements as a natural input method.

What it does

MurMur is a silent interface for AI agents. Using two IMUs—one attached to the jaw and one as a head-motion reference—it recognizes jaw gestures in real time and maps them to actions like opening ChatGPT, generating an app, or checking the weather, all without touching a screen or speaking.

How we built it

We built MurMur using an Arduino UNO Q, two LSM6DSOX IMUs, and a custom machine learning pipeline. One IMU captures jaw motion while the other compensates for head movement. A classifier running on the UNO Q's Linux processor recognizes gestures and sends commands to a Python agent, which automates AI applications and browser actions using Playwright. The entire classifier was developed from scratch, including the dataset.

Challenges we ran into

The biggest challenges were separating intentional jaw gestures from normal head movement, collecting consistent training data, achieving reliable real-time classification, and integrating hardware, embedded software, machine learning, and browser automation into one seamless pipeline.

Accomplishments that we're proud of

We're proud of building a complete end-to-end system from custom hardware and embedded firmware to machine learning inference and AI agent control in just a hackathon. Watching a silent jaw gesture launch and control an AI agent felt like a glimpse of a new way to interact with computers.

What we learned

We learned how important sensor fusion, real-time inference, and low-latency system integration are for wearable interfaces. We also gained experience designing ML models that work reliably on noisy biosignal-inspired sensor data along with developing valid datasets.

What's next for MurMur

We're expanding beyond IMUs to richer sensing modalities like EMG to improve accuracy and enable a much larger vocabulary of silent commands. Our long-term vision is a lightweight wearable that becomes a universal, always-available interface for AI agents and everyday computing.

Built With

• arduino
• cpp
• hardware
• imu
• iot
• playwright
• python
• qualcomm
• scikit-learn