Nod

Inspiration

Being able to use the internet shouldn't be a luxury. For people with arthritis, cerebral palsy, ALS, or spinal cord injuries, even a basic task like reading the news or paying a bill can be exhausting, or impossible. The special hardware that's supposed to help can cost thousands of dollars, which puts a huge strain on families.

We wanted to help people with accessibility issues and make sure everyone has an equal chance to use the internet.

What It Does

Nod is a hands-free way to browse the web, built for people who have a hard time with fine motor skills. It removes the need to type or struggle with a standard mouse. Users can click exactly what they want, select buttons, and scroll up and down web pages entirely hands-free.

Instead of the fine hand movements a normal mouse needs, Nod uses head movements and voice commands. That means users can read the news, pay bills, shop, and chat with friends on their own, without the physical strain that usually comes with getting online.

Sensors: Camera + Microphone

Nod's whole input system runs on sensor data from hardware people already own.

Camera: a standard webcam turns small head movements into input. We use Google's MediaPipe Face Landmarker to map 478 face landmarks and work out head turn and tilt, then smooth it so the controls feel steady.
Microphone: the mic is our second sensor, capturing voice commands for hands-free actions like clicking and searching.

Combining two sensors makes Nod more flexible. Users can choose whichever input is more comfortable for them, or use both together.

Local Voice Processing (Python + SciPy)

Our voice system uses a fully local audio signal-processing pipeline built in Python with SciPy. Raw microphone data is processed on the user's own computer instead of being sent to a cloud service. That keeps latency low, works without relying on outside services, and protects user privacy, which matters a lot when the tool is used for everyday things like banking and messaging.

AI Image Lab: Digital Forensics on Vultr

Hands-free access also means being safe online. Our AI image lab helps users tell whether an image came from a real camera or an AI model, and it runs on a Vultr server.

Vultr: our Vultr server runs the digital forensics pipeline. Images are sent to the server, scanned there, and the verdict is sent back to the lab, so the heavy analysis doesn't slow down the user's own computer.
Digital forensics: the server scans the image data, including EXIF metadata and other signals, to detect whether the image came from a real photo or an AI model instead of just trusting how it looks.
Results show up right in the lab, with AI-detected images clearly labeled, so users can practice telling real from fake.

UI/UX

Accessibility tools often feel clinical, so we went the opposite direction. Nod has a calm, welcoming sky-themed interface with large click targets, clear visual feedback, scroll animations, and a light-to-dark theme where the day sky turns into a starry night with a moon. It's built to feel friendly from the first click.

Social Good

Nod gives people their independence back. It runs on the basic computer parts people already own, so there's no expensive hardware and no financial barrier. A physical disability shouldn't keep anyone out of the digital world.

How We Built It

Front end: HTML, CSS, and JavaScript with custom cloud artwork, animations, and a day/night theme
Camera sensor: webcam input run through Google's MediaPipe Face Landmarker (478 face landmarks) to calculate head turn and tilt, with smoothing for steady control
Microphone sensor: voice input handled by a fully local Python + SciPy pipeline
AI image lab: a Vultr server running digital forensics (EXIF and other image-data analysis) to detect real vs. AI-generated images
Practice Tools: safe pages where users can try Nod's controls before browsing for real

What We Learned

Accessibility is a design problem as much as an engineering one. Big targets and clear feedback matter as much as accurate tracking.
Raw sensor data is messy. Even with 478 landmarks, head tracking looks shaky until you smooth it, and mic input needs filtering to be reliable.
Running processing locally protects privacy, while a cloud server like Vultr is great for heavier jobs like image forensics.

Challenges We Faced

Turning face landmarks into steady head turn and tilt input without jitter
Cleaning up microphone noise and detecting voice commands reliably with SciPy
Getting the forensics to tell AI-generated images apart from real ones
Setting up and deploying the forensics server on Vultr
Coordinating a three-person team and polishing the design under hackathon time pressure

What's Next

We want to keep improving Nod's accuracy, add more voice commands, expand the forensics tools, and test with real users who face these challenges every day. Our goal is simple: a free, affordable way for anyone to get online easily.