nourish

Inspiration

Someone on our team was trying to eat healthier but quickly realized that those recipes they saw on YouTube reels use ingredients that are expensive or hard to source in Indonesia. Now, imagine what it's like to be someone who has more urgency to watch their diet...say, a pregnant woman? A quick google search reveals that children of mothers with third trimester anemia are 7.18x more likely to be stunted.

Problem -> Feature

Pregnant women have no practical way of tracking the essential nutrients they need -> A hassle-free tracker that logs your nutritional intake using a single photo
Pregnant women need to eat healthy, but foods marketed as healthy tend to be more expensive and western-centric -> A menu planner that gives you recipes tailored to your personal daily intake requirements, using cheap and locally abundant ingredients
Doctors only get a snapshot of pregnant women's diet, making it hard to estimate if they have fulfilled their nutritional needs -> History of logs and an analytical dashboard that shows them exactly what their patient has been eating

How we built it

Photo tracker: Vision + text models infer the ingredients used for the dish and the nutritional value of the meal, specifically protein, iron, folate, and calcium
Menu planner: Text model analyzes the user's history of meals to find out what kind of dishes they like, and then generates a recipe using local ingredients based on the user's inputted location and average budget per meal
Analytics: Rules dictate if the user is significantly behind the day's target intake. If the user has not met the target for 5+ consecutive days, a warning will be raised for the doctor to pay special attention to that issue.

Challenges we ran into

💼 Somebody still needing to do WFO on Thursday and attend WFH meetings on Friday
🏍️ Somebody needing to go on an 8-hour motorbike ride to get to UMN
🔀 Pivoting ideas a million times
🤰🏻Faking pregnancy for the video

Accomplishments that we're proud of

Getting everything done in less than 24 hours
Showering more than we did in Garuda Hacks 6.0

What we learned

Gemini's vision model is really good and also free
Pollinations AI does not need any sign up or token or that sort of thing

What's next for nourish

Compatibility with SATUSEHAT
Expand to other conditions e.g. diabetes, hypertension

Disclaimer of AI Use

Claude, ChatGPT: Brainstorming, crawling for data, syntax lookup, debugging, boilerplate generation
Gemini (specifically gemini-3.1-flash-lite): Vision model to understand user's input photo and menu generation
Pollinations AI: Generation of photos for menu recommendation
Antigravity: AI-powered code editor