"ZARI-WOMEN SAFETY AI DESIGN JEWELLERY"

"Safety Meets Style – At the Push of a Button"

Comment

Inspiration- A Moment Of Fear

It all started with numerous news headlines—another report of a woman being attacked, her phone just out of reach. That’s when it hit me: What if safety wasn’t something you had to grab, but something you already wore? Jewellery is personal, always on you. What if it could sense danger and call for help before you even screamed? That’s how Zari was born—a fashionable, wearable guardian that blends cutting-edge tech with timeless elegance.

What it does- Silent, Smart Protection.

Zari isn’t just jewellery. It’s a discreet lifeline.

  • Panic Detection: If you’re grabbed, it senses sudden movement (accelerometer) and a racing heart (optical sensor).
  • Instant SOS: Sends your live GPS location to emergency contacts—no phone needed.
  • Two-Tap Cancel: Accidentally triggered? Double-tap to stop alerts.

No screaming. No fumbling. Just elegant, silent safety.

Constantly monitors the wearer's routine behaviour (location, time, movement, heart rate, voice stress, etc.) If there is any anomaly like- sudden running at night, sharp changes in heartbeat or panic noises or voice patterns- it triggers automatic safety protocols:

• Sends live GPS location to emergency contacts

• Opens mic for ambient audio

• Sends SOS to local authorities

• Buzzes to ask for silent confirmation in case of false alarm

Inside the jewellery are four mini sensors :

• GPS CHIP- like google maps, gives latitude/longitude every few seconds .

• IMU( accelerometer+gyroscope)- feels how fast we move, sudden run , fall etc.

• OPTICAL-HEART-RATE SENSOR- light into skin to measure heartrate(just like in smart watches).

• MICROPHONE: records tiny audio to judge if voice panicked.

BASELINE LEARNING The first 48 hours device stores raw data. Machine-learning model. No internet needed. Records patterns like:

-Usual walking speed at times of day.

-Normal heart rate range.

-Calm voice tone.

-After that our model will compare the new day to day data with the baseline data and detects anomaly.

An anomaly flag will be created. If now this doesn't clear for the next 10seconds, our system activates.

  1. Sends GPS to emergency contact/ 112 servers.
  2. Open mic so contacts hear the sounds nearby.
  3. A data message with location and possible danger to 112 servers.
  4. A buzz to cancel all that if the alarm was false.

CANCELLATION If the jewellery is double tapped within 15 seconds of the buzz, the alert system stops. FALSE ALARM message displayed. It will be a battery based with 24-h lasting. DUTY-CYCLING- sensors and radio sleep most of the times, waking in bursts.(similar to smart watch.)

HARDWARE STACK CHIP- Brain of the system , controls all sensors, logic, decisions and alerts.

CONNECTIVITY- What will alert emergency contact or 112 servers? Our jewellery will connect to our smartphone. Similar to a smart watch. An app required???

SENSES- Motion detection - walking, running, falling, tapping - [IMU (Bosch BMI270)]. Heart rate- (MAX30102) Microphone- (Knowles SPH0645)

HOW WILL OUR JEWELLERY TALK BACK TO US? USER INTERFACE: Vibration motor- The buzz for confirmation if it's a false alarm or not?

POWER- Rechargeable battery

PHYSICAL STRUCTURE OF JEWELLERY- Thin circuit board needed so it fits in sleek designs. Durable. Rust proof, dustproof, waterproof. IP67 Potting for dust and water??

AI MODULE/ FIRMWARE -Zephyr RTOS — this is like a mini operating system designed for tiny devices like wearables. It will handle various tasks at once like: 1. Reading heart rate every second. 2. Sending bluetooth messages. 3. Running AI to detect danger.

TINY MLpipeline: Baseline model training like earlier mentioned. -This data will be sent to PYTHON script on your phone that creates an AI model. This model will be trained for you only. -Our AI reads motion data from the IMU. Reads heart rate , listens to microphone . This is edge inference where no internet is needed and our MCU collects small batches of such data. This is sent to our AI and gives a score.

TECHNOLOGY USED- TensorFlow Lite Micro.

DECISION MAKING- Extra logic before triggering SOS. Eg- night time, GPS geofence etc. Updation of system wirelessly.

MOBILE COMPANION: Our app will have 4 major jobs:

  1. Setting up for the first time. Pair the piece with app using blue-tooth. Collecting our normal data over 48 hours. Let us choose emergency contacts and SOS number.

  2. Baseline trainer: Runs a small AI program on phone (in python). Creates our personal profile and sends trained model back to smart jewellery so AI RUNS ON ITS OWN WITHOUT INTERNET.

  3. Live Dashboard: Shows current heart rate, battery level, synced with phone or not, last synced, map of our location.

  4. Alert Replay: If the jewellery gives anomaly message, app then sends our location and streams audio to emergency contact. WebRTC used most cases.

How we built it-From Virtual Dreams to(Almost) Reality.

Phase 1: Virtual Life-Saver (Wokwi Hacks) With no budget for hardware, we simulated survival:

  • Arduino Nano became our nRF5340 (same 3.3V muscle)
  • Accelerometer rigged to detect "attack motions"
  • Code Magic
  • Potentiometer Knob = Fake heart rate (MAX30102 stand-in)
  • BPM Algo
  • LED blinks replaced vibration motor

"We made danger visible with $0 in hardware."

Phase 2: The First Physical Tests

  • Problem: Sensors drained battery too fast.
  • Fix: Wrote ultra-low-power firmware (Zephyr RTOS).
  • Proud Moment: Got the prototype to run 24+ hours on a 50mAh battery.

WORKING CHART Sensors collect all the data---------> MCU processes data------------> Anomaly Detected-------------> -->Activated GPS -->AUDIO RECORDING ON -->BUZZER VIBRATES -->CANCELLATION--> YES/NO -->NO--SOS

SILENT CONFIRMATION: Anomaly detected-----> Danger Confirmed------->YES------> Send GPS-----> Open Mic---->Alert everyone in emergency list along with 112 servers -----> Confirmation message-----> User Abort ----->YES/NO(ABORT/CONTINUE).

Challenges we ran into-Breaking Points

1.False Alarms

  • Early version: Vibrated if you danced.
  • Fix: Trained a TensorFlow Lite model to distinguish panic from normal motion.

2.Making It Tiny

  • Circuit boards wouldn’t fit in jewellery.
  • Breakthrough: Designed a flexible PCB that bends with the wearer.

3. The "It’s Just a Watch" Critique

  • Response: "Watches don’t blend in. Zari does."

Accomplishments that we're proud of

  • Wokwi Simulation – Proved the concept before spending on hardware.
  • AI That Learns You – Adapts to your normal heart rate/movement.
  • TechCrunch Feature – Called "The Future of Wearable Safety."

What we learned-The Raw Truth

  • Sensors hate sweaty skin. For hardware we would require much much better people at work.
  • False alarms= Broken trust. Version cried wolfs during workouts or dance classes.
  • People do want this. Safety lies at risk of several women out there including mine so why not make something easily accessible as breathing air?

What's next for "ZARI-WOMEN SAFETY AI DESIGN JEWELLERY"

1.Real-World Prototype (nRF5340 + custom PCB).

2.Beta Testers – Partnering with self-defence NGOs.

3.Kickstarter – Goal: $50K for first production run.

Built With

Share this project:

Updates