Teman Gizi

Inspiration

Indonesia is facing a growing public health crisis driven by diet-related illnesses. While our local cuisine is flavorful and culturally rich, it often lacks standardized nutritional information—making it hard for people to make healthy decisions. Inspired by the rise of hypertension, diabetes, gout, and obesity in Indonesia (Kemenkes RI, 2023), we built Teman Gizi to help users identify hidden health risks in the food they eat every day.

What it does

Teman Gizi is a mobile app that allows users to scan Indonesian foods using their phone camera and instantly see estimated nutritional values like calories, fat, sugar, and sodium. The app helps users identify which foods may increase their health risks and provides visual warnings for foods high in risk factors. It also offers simple suggestions for healthier alternatives or moderation tips.

How we built it

We trained a custom machine learning model for Indonesian food image recognition using a dataset curated from online food photos and field data. Once a food is recognized, the app retrieves estimated nutritional content using a localized nutrition database built from the Indonesian Food Composition Table (Puslitbang Gizi, 2021) and recent academic sources. The app itself is built using Flutter for cross-platform use, with a REST API backend to support scalability and speed.

Challenges we ran into

1. Lack of open datasets: Most nutrition databases are Western-focused. We had to collect and clean local food data manually.
2. Food variation: The same dish may vary in ingredients and portion size across regions, making accurate estimation difficult.
3. Model confusion: Many foods look visually similar (e.g., soto vs. rawon), so we built a confidence rating system to handle uncertainty.
4. Estimating portion size from images remains an ongoing challenge.

Accomplishments that we're proud of

1. Successfully trained a working food recognition model specifically for Indonesian cuisine, something very rare in existing ML datasets.
2. Built a functioning app prototype that can recognize food, show nutrition facts, and offer recommendations—all within seconds.
3. Made health tech more culturally relevant and accessible for Indonesians.

What we learned

We learned that solving health issues through tech requires not just accuracy, but empathy—understanding local food culture, behavior, and accessibility. We also gained skills in data curation, model training, and UX design for behavior change. Most importantly, we learned how impactful tech can be when rooted in local problems.

What's next for Teman Gizi

1. Expand our food database with more regional dishes from across Indonesia.
2. Integrate portion size estimation using object detection and AR.
3. Add health profile personalization—e.g., alerts for diabetic users.
4. Partner with local healthcare providers, nutritionists, and government agencies to validate and distribute the app at scale.

Built With

• expo.io
• javascript
• node.js
• python
• railways
• react-native
• supabase
• tensor