Severn LS Edge AI

Teaching the Next Generation of AI — One Sensor at a Time

An open-source, privacy-first platform that empowers 4th and 5th grade students to build their own AI systems using real hardware — no cloud required, no data leaves the classroom.

Imagine teaching a 10-year-old that they can create their own "smart" device — one that recognizes gestures, detects movement, or knows when someone is nearby — and all of it happens right on a tiny computer they can hold in their hand.

That's what Severn LS Edge AI does.

We're building an open, accessible alternative to tools like Google's Teachable Machine, but with a critical difference: your data never leaves your device. Students learn that artificial intelligence doesn't require sending information to faraway servers. It can happen right here, right now, on the "edge" — on the device itself.

Most AI educational tools work like this:

1. Collect data (photos, sounds, movement) from students
2. Send it to cloud servers owned by large companies
3. Process it somewhere far away
4. Send results back

Students never see where their data goes. They don't understand who has access to it. They become comfortable with surveillance as the default.

Edge AI flips this model:

1. Collect data from sensors on the Arduino
2. Process it right on the device (or locally in the browser)
3. Train and run AI models without internet
4. Your data never leaves your control

This teaches students that privacy and intelligence can coexist.

• Computational Thinking: Breaking problems into sensor inputs and outputs
• Data Literacy: Understanding what sensors measure and why
• Critical Thinking About Technology: Asking "where does my data go?"
• Hardware Fundamentals: Connecting physical devices to software
• Scientific Method: Collecting data, making predictions, testing hypotheses

┌─────────────────────┐      Bluetooth      ┌─────────────────────┐
│                     │    ~~~~~~~~~~~~>    │                     │
│  Arduino BLE Sense  │                     │   Your Computer's   │
│                     │    <~~~~~~~~~~~~    │      Web Browser    │
│  - Motion sensor    │                     │                     │
│  - Proximity sensor │                     │   Shows live data!  │
│                     │                     │   Trains AI models! │
└─────────────────────┘                     └─────────────────────┘
         ^                                           ^
         │                                           │
    Feels motion                                Runs locally
    Detects closeness                           No internet needed
    Runs AI models                              Your data stays here

Hardware Layer:

• Arduino Nano 33 BLE Sense Rev2
• BMI270 IMU (accelerometer/gyroscope) + BMM150 (magnetometer)
• APDS9960 (proximity/gesture/color sensor)
• Onboard BLE radio for wireless communication

Firmware Layer (arduino/severn_ble_sense/severn_ble_sense.ino):

• Reads accelerometer at 50Hz (every 20ms)
• Reads proximity sensor at 10Hz (every 100ms)
• Exposes data via two BLE GATT services:
  • Accelerometer Service (UUID: 00001101-...): 3 floats (X/Y/Z) = 12 bytes
  • Proximity Service (UUID: 00001102-...): 1 byte (0-255)
• Generates unique device names from MAC address (critical for classrooms)

Web Layer (index.html):

• Pure HTML/CSS/JavaScript (no frameworks, no build step)
• Uses Web Bluetooth API for device discovery and connection
• Real-time sensor visualization
• Works offline — just open the file in Chrome/Edge/Opera

ML Layer (Integration with Edge Impulse):

• Export sensor data for model training
• Deploy trained TinyML models back to Arduino
• Run inference entirely on-device

What You Need:

• Arduino Nano 33 BLE Sense Rev2 boards (one per student or pair)
• USB cables for initial programming
• Computers with Chrome, Edge, or Opera browsers
• Labels/stickers for device identification

One-Time Setup (per device):

1. Install Arduino IDE (version 2.x recommended)

2. Install the board package:

   • Tools → Board → Boards Manager
   • Search "Arduino Mbed OS Nano Boards"
   • Install

3. Install required libraries (Sketch → Include Library → Manage Libraries):

   • ArduinoBLE
   • Arduino_BMI270_BMM150 (for Rev2 boards)
   • Arduino_APDS9960

4. Upload the firmware:

   • Open arduino/severn_ble_sense/severn_ble_sense.ino
   • Select board: Tools → Board → Arduino Nano 33 BLE
   • Select port: Tools → Port → [your Arduino]
   • Click Upload

5. Label your device:

   • Open Serial Monitor (9600 baud)
   • Note the unique device name (e.g., "Arduino-A1B2")
   • Write this on a physical label and attach to the Arduino

6. Repeat for all devices

1. Make sure your Arduino is powered (via USB or battery)
2. Open the web page in Chrome, Edge, or Opera
3. Click "Connect to Arduino"
4. Find YOUR device in the list (look for the name on your label!)
5. Click Pair
6. Watch your sensor data appear in real-time!

Critical Questions We Encourage:

• "When I use an app, where does my information go?"
• "Who can see what I'm doing?"
• "Can I make something smart without sharing my data?"
• "Why do some companies want my data?"

The Answer Edge AI Provides:

• Intelligence doesn't require surveillance
• You can build powerful AI systems that respect privacy
• The future of technology can be different

Google's Teachable Machine is fantastic for quick demos. But for real education, we need more:

We want students to be creators, not just consumers.

Arduino Nano 33 BLE Sense
├── BLE Peripheral Role
│   ├── Device Name: "Arduino-XXXX" (unique per device)
│   │
│   ├── Accelerometer Service (0x1101)
│   │   └── Accelerometer Characteristic (0x2101)
│   │       ├── Properties: READ, NOTIFY
│   │       └── Value: [float32 X][float32 Y][float32 Z] (12 bytes, little-endian)
│   │
│   └── Proximity Service (0x1102)
│       └── Proximity Characteristic (0x2102)
│           ├── Properties: READ, NOTIFY
│           └── Value: [uint8] (1 byte, 0=far, 255=near)

Web Browser (Central Role)
├── Web Bluetooth API
│   ├── Scans for devices with name prefix "Arduino"
│   ├── Connects to GATT server
│   ├── Subscribes to characteristic notifications
│   └── Parses binary data and updates UI

Accelerometer packet (12 bytes):

Byte 0-3:   Float32 X (little-endian)
Byte 4-7:   Float32 Y (little-endian)
Byte 8-11:  Float32 Z (little-endian)

Proximity packet (1 byte):

Byte 0:     Uint8 (0-255)

• Arduino firmware with accelerometer and proximity
• Unique device naming for classroom management
• Web Bluetooth connection and real-time display
• Cross-browser support (Chrome, Edge, Opera)
• Mobile-responsive design

• Data recording and export (CSV, JSON)
• Edge Impulse integration for ML model training
• On-device inference deployment
• Activity recognition demo (walking, running, jumping)
• Gesture recognition demo (wave, circle, shake)
• Classroom dashboard for teachers
• Lesson plans and curriculum materials
• Multi-language support

• Arduino Nano 33 BLE Sense Rev2 (recommended)
• USB-C cable for programming
• Optional: Battery pack for wireless operation

• Arduino IDE 2.x
• Modern web browser with Web Bluetooth support:
  • Chrome 56+
  • Edge 79+
  • Opera 43+
  • Note: Firefox and Safari do not support Web Bluetooth

• ArduinoBLE (by Arduino)
• Arduino_BMI270_BMM150 (by Arduino) — for Rev2
• Arduino_APDS9960 (by Arduino)

Libraries won't install:

• Install the board package first (Arduino Mbed OS Nano Boards)
• Use Arduino IDE 2.x for best compatibility

Upload fails:

• Double-press the reset button to enter bootloader mode
• Try a different USB cable (some are power-only)
• Check that the correct board and port are selected

Sensors not initializing:

• Check Serial Monitor at 9600 baud for error messages
• Ensure you're using the correct libraries for Rev2 (BMI270, not LSM9DS1)

"Web Bluetooth not supported":

• Use Chrome, Edge, or Opera
• On Chrome, check chrome://flags for Bluetooth settings

Can't find device:

• Make sure Arduino is powered on (LED visible)
• Check Serial Monitor to confirm "BLE advertising started"
• Move closer to the device (within 10 meters)

Connection drops frequently:

• Check for BLE interference (other devices, USB 3.0 hubs)
• Ensure Arduino has stable power

severn/
├── README.md                    # This file
├── CLAUDE.md                    # Context for AI assistants
├── index.html                   # Web interface (single file, no build)
└── arduino/
    ├── README.md               # Firmware setup guide
    └── severn_ble_sense/
        └── severn_ble_sense.ino # Arduino firmware

We welcome contributions from educators, developers, and students!

• Educators: Test in your classroom and share feedback
• Developers: Improve the web interface or firmware
• Students: Report bugs and suggest features
• Everyone: Help translate materials for other languages

1. Clone the repository
2. Upload firmware to your Arduino
3. Open index.html in a supported browser
4. That's it! No build process, no dependencies

We believe the next generation deserves to understand AI, not just consume it. They should know:

1. AI is not magic. It's math, sensors, and code — things they can learn.
2. AI doesn't require big companies. You can build intelligent systems yourself.
3. Privacy matters. Smart technology can respect your data.
4. Hardware is accessible. A $30 Arduino can do amazing things.
5. Open source empowers. When you can see the code, you can change the code.

Google's Teachable Machine democratized ML demos. But it's still:

• A black box (students can't see how it works)
• Cloud-dependent (data leaves your device)
• Limited to camera/microphone (no hardware sensors)
• Not deployable (you can't put it on your own device)

Severn LS Edge AI fills this gap by giving students real hardware, real code, real privacy, and real understanding.

This project is open source. See LICENSE file for details.

Built for the students of Severn Lower School, inspired by educators everywhere who believe kids deserve to understand the technology shaping their world.

Teaching Edge AI. Protecting Privacy. Empowering the Next Generation.