CradleGuard

Main Menu
Live feed
Customization

Inspiration

For us, this project is very personal. We all have younger siblings, so we’ve all watched our parents’ stress levels spike the moment a baby cries. We’ve seen the anxiety that comes with leaving a child alone, even for a minute, to the point where true relaxation feels impossible. Existing baby monitors fall short because they’re passive. They stream video, but they don’t understand what they’re seeing. The inspiration behind this project was that we wanted to transform passive monitoring into active intervention. CradleGuard is our answer: an intelligent system that doesn’t just show you a crying baby, it identifies dangerous situations before they happen, giving parents a second pair of eyes that never blink.

What it does

CradleGuard is an AI-powered safety system that combines a Raspberry Pi with the Gemini 2.5 Flash API to proactively protect infants from household hazards.

Smart Room Scanning: During setup, CradleGuard scans the camera view and automatically identifies standard hazards, like exposed outlets, sharp objects, or choking hazards, using . Parents can then manually add any additional items the scan might have missed.
Personalized Safety Rules: The Gemini API analyzes the detected hazards and generates context-aware risk scenarios unique to that room. Parents can also input custom rules, such as "alert me if the baby approaches the floor lamp".
Alert Settings: CradleGuard triggers an alarm on the device and instantly sends an alert to the parent’s app, complete with a live view of the camera. Parents can choose whether to receive an alert on the website, on the speaker, or on both when the baby is in danger.
Real-Time Live Camera Feed: The system continuously monitors the live camera feed using a reasoning loop (YOLO-World) to detect if the baby is interacting with a hazard. If a risk is identified, Gemini verifies this hazard and alerts the parents immediately.

How we built it

Frontend: Typescript, React, Vite, Tailwind CSS
Backend: Python
Danger Detection: Gemini 2.5 Fast API, FastAPI, WebSocket, YOLO-World
Hardware: Raspberry Pi
Design: Figma, Canva

Challenges we ran into

We wanted a concept that was both ambitious and personally meaningful, settling on the right idea took a lot of time.
Merging code across 4 member team led to some intense merge conflicts, but it taught us how to resolve such conflicts.
We wanted the danger detection to be as accurate as possible, so it was difficult choosing the right detection tools. We ultimately combined YOLO-World for setup-time scanning and Gemini for real-time reasoning.
With a 36-hour timeline, we had to make tough calls. Some features were cut due to limited time, but we stayed focused on delivering the functions that will make a meaningful difference.

Accomplishments that we're proud of

We built a solution that doesn’t just monitor, but actually prevents accidents AND relieves parents' stress. This feels genuinely meaningful to our team.
Our team collaborated so well despite the time pressure, and we’re proud of how we supported each other through the tough moments.
We couldn't connect the Raspberry Pi to Wi-Fi at first and spent 5+ hours trying to connect it, but we were able to solve this problem.

What we learned

For many of us, this was our first time working with a Raspberry Pi and YOLO-World. We learned how to work with these new tools.
We learned how to manage our time efficiently and establish a smooth workflow for collaboration

What's next for CradleGuard

Multi-camera support, installing multiple units and combining the screens for full-room coverage and more accurate detection.
Expanding to other age groups, such as adapting the system for toddlers or even pets.
Moving from web to iOS/Android for faster alerts and a smoother user experience.
Stop the baby with voice cloning by using ElevenLabs to clone a parent’s voice, so the system can speak soothingly to the baby when they're about to head into danger.