cocare ai

the problem

we were inspired by organizations like mariam's footsteps, which work with children with disabilities who often struggle to express their needs and feelings. caregivers carry a lot of emotional labour observing, logging and personalizing care, often without knowing how a child truly feels. we wanted something that respects every child's dignity while easing that load.

what it does

• privacy-safe edge sensing detects play, rest and agitation without storing raw video or audio.
• smart summaries every 30-60 minutes that caregivers can quickly edit or approve.
• nonverbal children can express themselves through tablet buttons or eye-tracking, folded straight into the care log.
• calming sensory games that help the child relax and engage.

how we built it

• mediapipe tracks 33 body landmarks for real-time posture and motion.
• the gemini api translates movement patterns into plain summaries - "sitting calmly", "jumping", signs of overstimulation - and the caregiver side suggests responsive activities.
• python + opencv for frame processing, with threading so the backend stays smooth while gemini is queried asynchronously.
• a react + tailwind + typescript front end, json logs, and a streamlit prototype.

challenges

• balancing real-time accuracy with usability.
• earning privacy and trust when working with vulnerable children.

what i learned

• privacy is non-negotiable with vulnerable populations.
• small design details empower - even a simple 'i liked this' button can make a child feel heard.
• trust comes from transparency and control, which is why caregivers edit the summaries instead of just receiving them.

team

built with jai pannu, hari mallampalli and shreay patel.