Idea: How to use Google AI for the most impact? Can we make a difference in the lives of millions of needy people?

Our idea promotes creative and responsible use of the power of Google AI for a significant impact on the lives of people who will greatly benefit from these technological advances

Hybrid AI: client and server side While healthcare data is multimodal in nature, it suits to utilize the unique strengths of Multimodal Gemini Models and long context

Offline/Client AI: Given the privacy needs of healthcare usecase, and the child-related photography rights, and the digital divide, the use of offline client AI lends here

Feature request for capabilities Client AI

We believe in power of offline AI but that if multimodal support can be enabled in client AI, it would make a lots of sense for helping the world make process on UN Sustainable Development Goals, especially in gender and health and education in the challenges (privacy needs of healthcare usecase, and the child-related photography rights, and the digital divide).

Given large MLLMs cannot be run on the client side, it would make sense for the Google team to consider how to bring multimodal support in client AI. Ideas could be how to use MediaPipe and small LLM with some finetuning to extract visual features from MediaPipe (posedetection or object detection) and then feed that visual features into a text LLM. This can happen on client side. It is possible to fine tune such combo of visual features with text for radical transformation of multimodal support on client AI

What are 3 challenges addressed by your solution?

(Health of millions of Girl children) : How to prevent girl child deaths in least developed countries?
(Mental health of millions of girl children with disability) : How to build hope for girls who have experienced violence and disability?
(STEM and Medical Education): How to transform STEM education for girls? How to build medical capacity in Africa to improve maternal health in pregnancy and child birth?

What are the issues addressed by this project

Does your project solve a significant issue? How did you arrive at the 3 problem statements?

To identify most pressing and significant issues, we research extensively with many reports from UN Women, UNICEF, WHO. So, we identified the 3 areas of contribution for maximal impact. Our project aligns with ambitions of UNICEF Strategic Plan 2025, and UNICEF Gender Action Plan 2025.
What are the issues addressed by this project

You mentioned contributions of 3 themes. What is your Contribution #1?

Impact & Relevance (Contributing Theme #1):

Specific Problem:

Many girls die in their early years in certain countries due to a lack of attention to health. To understand this problem, we queried the UNICEF database for Girl Child Girl Mortality Rate (under 5 years).
Our finding is at least 1 in 20 girls die before their 5th birthday in certain countries (the chart below shows the specific countries).

Insight on why this Problem?

Lack of access to healthcare services in the least developed regions affects the health of girl children and her survival as per UNICEF's Report 2023 on "Health and Survival" (Ref 4). If the mother is a young adolescent (<19 years), she is dependent upon the male gender to access healthcare facilities due to gender balance in families, as per UN data (Ref 5).
Detecting the 8 danger signs in infants early enough can prevent the death of the girl.
Early detection of danger signs is critical but is often missed out due to lack of trained doctor in rural areas.

Innovation & Creativity (Contributing Theme #1):

What is the Solution (Contributing Theme #1)?

Multimodal AI is used for Early detection of danger signs of infant health.
The AI alerts the family if the child is in danger, and also alerts the nearby health volunteer.

Demo of AI Solution #1?

Demo shows how GEMINI Pro is able to detect danger signs in infant@home and advises family members to take the infant to the clinic.
We demonstrate that GEMINI can reason over complex multimodal input. Demo shows GEMINI can detect danger signs in the breathing patterns of infants by reasoning over Few Shot Learning with Contrastive examples of healthy and non-healthy infants

Solution 1

Innovation: Does the project offer a new and innovative approach to the chosen issue? (Contributing Theme #1)

While literature has solved Deep Learning image classification problems, no one has solved the problem of video classification of new born breathing patterns. For the first time, we create a AI that identifies danger signs in child's breathing patterns. We feed animated GIF to GEMINI and also give Few Shot Contrastive Video Examples of healthy & unhealthy breathing videos. Thus, we were able to reason over child's breathing movement patterns such as fast chest indrawing, which can be result in loss of life.
Solution 1

Innovation: Does it leverage AI in a creative and effective way? (Contributing Theme #1)

Thus girl child deaths in African homes can be prevented by early detection using Multimodal Instruction Tunning with video and Animated GIFs of breathing patterns. The lack of doctors in Africa has caused a child to die every few minutes. By enabling mothers to use the intelligence of GEMINI to reason over multimodal data of the child, the project enables early detection of danger signs.

Impact: Does this project have Potential for Positive Impact? (Contributing Theme #1)

3 million girl children can be saved from deaths each year.

Solution 1

Impact: Does it addresses a significant need? (Contributing Theme #1)

Solution 1

We aligned the problem statement for this hackathon with the UNICEF Strategic Gender Action Plan 2025. This project contribution #1 addressed the key challenged mentioned as Goal Area #1 in the UNICEF Strategic Plan report. (screenshot below shows how we solved a problem that is in alignment with the ambitions as mentioned in UNICEF Strategy)

Solution 1

Presentation: Does the demo video demonstrate the tool adequately? (Contributing Theme #1)

The 3 min video only covers a glimpse of the demos .
For adequate video recording of the demos, Please refer the 7 min video demo in the project website.

You mentioned contributions of 3 themes. What is your Contribution #2?

What is this theme of contribution about?

This theme is about building hope for girls, who have lost hope after experiencing disability or violence. The idea is to employ AI to offer First aid for building hope, when all hope is lost. For example, girls who have faced prolonged war experience mental trauma after the event. To help their mind to visualize new possibilities, this AI solution generates content to help them visualize a positive future. We employ AI to ensure content safety.

Solution 1

Innovation: Does the project offer a new and innovative approach to the chosen issue? (Contributing Theme #2)

The solution provides first aid (phycological first aid) to help traumatized young girls to visualize a positive future.

Solution 1

Innovation: Does it leverage AI in a creative and effective way? (Contributing Theme #2)

We employ stable diffusion and GEMINI to compete and iterate (inspired by Ian Goodfellow's GAN) to develop safe and effective content for responsible usage.

Solution 1

Impact: Does it addresses a significant need? (Contributing Theme #2)

Solution 1

Impact: Is there alignment with Women's Empowerment? (Contributing Theme #2)

Solution 1

Presentation: Does the demo video demonstrate the tool adequately? (Contributing Theme #2)

 The  video covers the demo .

You mentioned contributions of 3 themes. What is your Contribution #3?

What this contribution is about? (Contributing Theme #3)

Transforming STEM education for girls by creating AI for experiential learning (Impact: To enable girls with equitable quality education in STEM subjects, and enable girl to become change makers and leaders in medical studies)
Transforming Medical education by creating AI that can coach. (Impact: To coach health volunteers in countries where medical staff are not staffed enough)

Solution 1