Student Coordinator, CYPHERMIND
A reverse prompt engineering challenge at the Zephyrus 8.0 technological summit, run by the Department of Computer Science. I co-led it with Gopika Ravindran, from the first plan to the day itself.
Computer science student in Thrissur, Kerala, building computer vision and machine learning systems you control with your eyes and hands.
PythonOpenCVMediaPipePygameESP32ReactFlaskMySQLCTailwindDocker, exploringNeural networks, exploring
What I’m working on.
AIRA
MYSTContactless apnea alerts for newborns, from a camera. It counts every breath and raises the alarm after 20 seconds without one. No electrodes on fragile skin.
Try the monitorType with your eyes, select with a wink. Webcam only.
Try the keyboard ARIAPinch the air to switch a fan and a bulb through an ESP32.
Try the simulation QuorraDashboards for everyone who runs a college event.
Explore the rolesLight Link
A message sent across the room as blinking light, found and decoded by a webcam. No Wi-Fi, no Bluetooth. Next: ESP32 and a laser.
Try the demo
Hi, I’m Diuthendrh.
I’m a computer science student at Christ College, Irinjalakuda, and I’m curious about a lot of things. Coding is how I chase that curiosity, lately into computer vision and machine learning. Art is simply something I love.
Working with: Python, OpenCV, MediaPipe, Pygame, React, Flask, MySQL, C. Exploring: Docker and neural networks.
- Mindset
- Explore fearlessly. Think differently. Create relentlessly. Learn endlessly.
- Hobbies
- Drawing, music, and curiosity.
- Motto
- Jack of all trades, master of none, but oftentimes better than a master of one.
- Based in
- Thrissur, Kerala, India. Christ College (Autonomous), Irinjalakuda.
MYST
Type with your eyes. Select with a wink.
Multi-Sensory Yielding Support Technology is a virtual keyboard for people with paralysis and motor disabilities. It needs only a standard webcam: no keyboard, mouse, touchscreen or special hardware.
Highlight jumps: 0.0/s
Gaze navigation
Horizontal gaze slides a highlight across the keys. Normalized iris tracking, EMA smoothing, a 7-frame majority vote and gaze locking keep it steady.
Wink to select
A deliberate wink picks the key. MYST tells it from a natural blink by comparing the blendshape scores of both eyes.
Help while typing
Bigram word prediction on the Brown corpus, autocorrect and text-to-speech, across QWERTY and symbol layouts.
What didn’t work
Before splitting navigation from selection, we tried pure gaze selection six ways. None held up on a webcam in real use.
- Sclera detection
- Iris-to-eyelid ratios
- Eyelid distance
- Vertical blendshapes
- Real-time calibrationSmall errors added up until the cursor drifted.
- Full 2D gaze estimationFine in controlled light, not in real use.
After 13+ versions: gaze navigates, a wink selects.
Specs
- Built with
- Python, OpenCV, MediaPipe FaceLandmarker, Pygame
- Tracks
- 478 face landmarks and 52 blendshapes
- Hardware
- A standard laptop webcam
- Interface
- 60 FPS Pygame UI, live settings panel, saved calibration
- Iterations
- 13+ versions
ARIA
Pinch the air to switch a fan and a bulb.
ARIA reads your hand through a webcam, draws a glass interface you point at, and sends commands over WiFi to an ESP32 that switches two relays. No touch, no remote.
- Click a device to send a command.
Index fingertip
Your fingertip is the cursor. Hover over a device to target it.
Thumb and index pinch
A click. Pinch over a device to switch it on or off.
Index and middle pinch
Grab a device and drag it anywhere on the screen.
How it’s wired
- LaptopWebcam, MediaPipe hand tracking, OpenCV interface
- WiFiTCP connection to the ESP32 on port 8080
- ESP32GPIO 23 to relay IN1, GPIO 22 to relay IN2
- RelaysIN1 switches the fan, IN2 switches the bulb
Protocol
| Device | Laptop sends | ESP32 replies |
|---|---|---|
| Fan | LED:ON LED:OFF | STATE:LED:… |
| Bulb | RELAY:ON RELAY:OFF | STATE:RELAY:… |
The wire names stay LED and RELAY to match the firmware. A mock ESP32 script listens on the same port, so the whole loop runs on one laptop.
Quorra
One platform for the people who run college events.
Students, coordinators, managers and admins each get their own dashboard and their own API routes. Coordinators plan, managers approve, admins look after accounts.
A dashboard per role
Each role sees only its own tools, backed by its own set of API routes.
Forms and approvals
Coordinators file proposal, expense, venue, advertisement, volunteer and media forms. Managers approve venues and role requests.
Reports and alerts
Stats for managers, notifications for everyone, and event exports at the click of a button.
Event lifecycle
PendingApprovedUpcomingOngoingCompleted
Rejected
Example eventupcominghigh priority
Progress60%
Specs
- Frontend
- React 19, Vite, Tailwind, Framer Motion
- Backend
- Flask, SQLAlchemy, MySQL, bcrypt
- API
- 40+ endpoints for auth, events, forms, venues, notifications and role requests
- Roles
- Student, coordinator, manager, admin, spectator
Light Link
A message across the room, sent as blinking light.
A phone blinks its light. A laptop webcam finds it, follows it and decodes the message. No Wi-Fi, no Bluetooth: light is the only link. This is version 1 of an optical data link I’m building toward an ESP32 and a low-power laser that aims itself.
Try it: type a message, tap Send, and watch the receiver below decode it. The demo runs the real frame format in a simulated camera.
2 / 32
Bit rate
The light blinks for a few seconds first so the receiver can lock on.
SEARCHING0link quality
Bit ratemust match the sender
- FPS
- 30
- Valid
- 0
- Failed
- 0
- Messages appear here.
Verified or nothing
A message appears only if its CRC passes. Failed frames are logged as failed and their text is never shown.
Finds the blinking light
It looks for pixels that change over time, so a steady lamp is ignored, then follows the spot with a Kalman filter.
Same bit rate, both ends
Sender and receiver must agree, from 2 to 15 bits per second. Set it on the phone and on the receiver dashboard.
How it works
- PhoneBuilds a frame, Manchester-codes it, blinks it once with a short lead-in
- LightOn and off at 2 to 15 bits per second, 4 for a 30 fps webcam
- WebcamFinds the blinking spot, tracks it, reads its brightness every frame
- ReceiverRecovers the clock from the preamble, decodes, checks the CRC
Next
- ESP32A hardware-timed transmitter for steady bit timing
- LaserA low-power, eye-safe visible laser for range
- AimingThe receiver already outputs where the light is, ready to steer a pan-tilt mount
None of this is built yet. Today it is a phone, a webcam and Python on a laptop CPU.
AIRA
Contactless apnea alerts for newborns, from a camera.
Apnea & Infant Respiration Analysis. Newborns, especially preterm babies, can suddenly stop breathing. Hospitals watch them with electrodes on fragile skin. AIRA watches the chest through a camera instead and counts every breath in real time.
It warns at 15 seconds and alarms at 20 seconds without a breath, until a nurse confirms. One camera can watch several cots. When it can’t see clearly, it alerts staff instead of staying silent.
Try it: press Hold breath and watch the clock run to a warning, then an alarm.
Research prototype, not a medical device.
BREATHING40breaths / min
Time since last breath0.0 s
Breathing normally
Every breath, counted
Optical flow follows the chest rising and falling, and each rise is counted as a breath, live.
Warn at 15, alarm at 20
No breath for 15 seconds is a warning. At 20 seconds the alarm stays on until a nurse confirms it.
Never silently blind
When it can’t see clearly enough to count breaths, AIRA alerts staff instead of staying silent.
Results
On the public AIR-125 infant video dataset, with babies it was never tuned on.
- 88%of breaths detected
- 39 to 41breaths a minute counted when the experts counted 40
- 0false apnea alarms in 63 clips
- BetaHeart rate from skin colour (rPPG)
Next
- InfraredA night-vision camera so it works in darkness
- ThermalTo catch obstructive apnea, when the chest moves but no air flows
- Capacitive padA sealed, battery-powered pad under the mattress that senses breathing through the bedding, by electric field
Today it is built on a laptop and its webcam.