NurtureLink
A voice for every parent in their own voice
The heart of the problem
Every child runs toward a parent when they are scared, confused, hurt, sleepy, or unsafe.
But not every parent is able to answer at that exact moment.
Some parents are deaf or mute and may never get to call their child’s name aloud. Some parents are paralyzed, living with ALS, locked-in syndrome, or late-stage Parkinson’s. Their mind is fully present, but their body cannot respond. Some parents are working, away from home, or just in another room when something suddenly goes wrong.
The pain is different for each family. But the gap is the same.
The child needs the parent. The parent wants to respond. But the voice does not reach in time.
NurtureLink is built for that moment.
It helps parents protect, comfort, guide, and emotionally connect with their children using AI, computer vision, voice, and multimodal reasoning. Most importantly, every response is spoken in the parent’s own matched voice, so the child does not hear a machine.
They hear Mom. They hear Dad. They hear the person they trust.
What NurtureLink does
NurtureLink is a Gemma-powered assistive parenting app that gives every parent a way to stay emotionally and physically connected with their child.
It combines child safety monitoring, cry understanding, sign communication, bedtime storytelling, emotional check-ins, and parent voice synthesis into one system.
It is not just a safety camera. It is not just a baby cry classifier. It is not just a text-to-speech app.
It is a bridge between a parent and child when normal communication breaks.
The three features that matter most
1. GuardianWatch
An AI guardian that speaks in the parent’s voice
GuardianWatch watches over a child or elder through a webcam every 12 seconds.
Gemini 2.5 Flash analyzes each frame for possible danger such as:
sharp objects open flames climbing choking risk falls hand-on-chest cardiac signs unknown people unsafe surroundings
When danger is detected, GuardianWatch responds immediately.
First, it speaks calmly in the parent’s matched voice.
Example:
“Aarav, please put down the scissors. Mama is coming.”
Then it logs the incident with a thumbnail.
If danger is detected repeatedly, it escalates into an emergency mode with:
GPS coordinates one-click WhatsApp emergency message Gemini-written emergency call script local emergency number such as 108, 112, or 911
This is the most important part of NurtureLink.
A normal camera only records danger. NurtureLink responds to danger with the parent’s presence.
2. StoryWeaver
A bedtime story from a parent who may not be able to speak
StoryWeaver helps a parent create a personalized bedtime story for their child.
Gemma writes a four-scene story based on the child’s age, mood, and caregiver note. Each scene becomes a watercolor children’s-book illustration using Pollinations.ai. Then the story is narrated in the parent’s matched voice.
On screen, the parent’s photo gently pulses with sonar rings while the story plays.
For a working mother, it means she can still say good night. For a deaf parent, it means their child can hear a story in their presence. For a locked-in father, it means he can still put his child to sleep with a story created from his love.
This feature is not just about storytelling.
It is about preserving emotional closeness.
3. Parent Bridge
Understanding a baby’s cry and guiding the parent
Parent Bridge records 7 seconds of baby audio.
It first checks whether the sound is actually a baby cry using an Audio Spectrogram Transformer. Then a custom acoustic-feature classifier analyzes pitch, MFCC, spectral rolloff, zero-crossing rate, and other sound patterns.
It classifies the cry into one of five categories:
hunger gas tired burping discomfort
Then Gemma reasons over the cry result, baby photo, feeding context, diaper context, and caregiver note.
Instead of only saying:
“hunger”
NurtureLink gives a parent-like suggestion:
“Your baby may be hungry. The rhythmic cry pattern and timing after the last feed suggest offering milk first. Try feeding and watch if the cry settles.”
This matters because exhausted parents do not only need a label.
They need calm guidance.
Other features
Voice Setup
The parent records around 10 seconds of voice. NurtureLink matches it with one of 14 neural voices and creates a voice that feels close to the parent’s natural tone.
Average voice match: 98 percent.
EyeBridge
For paralyzed parents or users with limited movement, EyeBridge allows communication through head pose and iris dwell typing.
It uses MediaPipe Face Landmarker, One-Euro filter smoothing, and an inline ASL avatar through sign.mt.
This helps a parent type, choose phrases, and communicate without using hands.
SignSpeak
SignSpeak uses MediaPipe Hands to detect hand landmarks in real time.
It recognizes 18 rule-based signs and gives a response in around 2 seconds.
This helps bridge communication between sign language and spoken or written output.
EarBridge
EarBridge works in the reverse direction.
When a hearing child speaks, the app converts speech into captions, detects emotion, adds simple emoji meaning, and can show an ASL avatar.
This helps deaf parents understand not only what the child said, but also how the child feels.
CalmCue
CalmCue is a mic-only comfort system.
When the baby cries, it automatically plays a pre-saved soothing phrase in the parent’s matched voice.
For example:
“Shhh, it’s okay. Amma is here.”
The child hears comfort immediately, even before the parent reaches the room.
Child Voice Check-in
This feature helps non-verbal children express emotions visually.
The child can show how they feel through simple emotion choices, facial input, or visual check-ins.
It helps parents understand whether the child is happy, scared, tired, uncomfortable, or needs attention.
Architecture
Frontend
The frontend is built with:
React 19 TypeScript Vite Tailwind v4
It talks directly to Gemma and Gemini APIs using fetch and FormData.
All hand and face landmark detection runs inside the browser using MediaPipe Tasks Vision.
This means many assistive features work without sending private face or hand movement data to a server.
Backend
The backend is built with:
FastAPI PyTorch Hugging Face Transformers
It hosts google/gemma-3n-E2B-it for multimodal reasoning, story generation, sign support, child check-ins, and Parent Bridge diagnosis.
It also hosts the cry classifier pipeline.
Audio uploaded from the browser is converted from webm or opus to WAV using pydub and imageio-ffmpeg.
Microsoft Edge TTS is used for free neural voice synthesis across 14 voices.
Why Gemma matters
Gemma is important for NurtureLink because this problem is not only about classification.
A baby cry classifier can say “hunger.” A vision model can say “scissors detected.” A text model can write a story.
But a parent needs all of these to work together.
Gemma helps combine image, audio, and text context into one meaningful response.
It can reason like:
The baby is crying. The last feed was two hours ago. The baby’s face looks uncomfortable. The caregiver note says the baby usually cries before feeding. So the parent should try feeding first.
That is the difference between an AI tool and an AI parenting bridge.
Gemma also runs locally on a laptop using gemma-3n-E2B-it in float16 mode, fitting in around 5 GB RAM. This makes the system more private, cheaper to run, and less dependent on internet or API limits for core features.
Why this is different
Most assistive technology solves one problem for one group.
One app helps deaf users. Another helps paralyzed users. Another monitors babies. Another generates stories. Another does text-to-speech.
NurtureLink brings them together around one emotional truth:
A child should not lose the comfort of a parent’s voice just because the parent cannot speak, move, or be physically present.
That is why the app is designed around the parent’s voice.
Because for a child, safety is not only about receiving an alert.
Safety is hearing the voice they trust.
What we would build next
True voice cloning
Move from neural voice matching to true voice cloning using XTTS-v2 once it is stable on Windows and Python 3.12.
Forever Voice mode
Allow parents with terminal illness to record their voice, memories, advice, and answers to common questions while they still can.
Later, their child can continue hearing guidance and love in their parent’s voice.
This would not replace the parent. It would preserve a part of their presence.
On-device GuardianWatch model
Train a local hazard detection model so GuardianWatch does not depend on Gemini API quota.
This would make safety monitoring faster, cheaper, and more private.
Native mobile app
Use Capacitor to convert NurtureLink into a mobile app so parents can use it anywhere.
Final message
NurtureLink is not only an AI project.
It is a communication bridge.
It gives a deaf parent a voice. It gives a paralyzed parent a way to respond. It gives a working parent presence when they are away. It gives a scared child the sound they trust most.
A parent’s voice can calm a child faster than any notification.
NurtureLink makes sure that voice is still there when it matters.
Log in or sign up for Devpost to join the conversation.