COVID-19 has transformed life for everyone, changing the way we live, love, work and most importantly eat. With irregular food supplies, on-and-off lockdown, it becomes crucial to mainatain a healthy diet and eating habits within availability. The worst hit are the ones with poor immunity who, uncoincidentally, intersect with diabetes patients. And the hassle of having to manage daily diet along with the day-to-day struggles of the new lifestyle that we are forced to adopt is outright cumbersome. Our product solves the problem of having to manually manage diet charts and daily calorie intake. This is particularly appealing to users who are watching their waistline, building muscle mass, trying to lose weight or battling diabetes.
What it does
Our baffling AI system can track your daily calorie intake with nothing but a picture of the food you are eating right now. It can also provide you with foods that you can make at home with the ingredients that you have, and the closest healthy food to any chosen food that neither compromise in taste nor nutrition. You take a picture, we will manage the rest.
How I built it
We use PyTorch in two instances:
- Onboard PyTorch model that can detect food class from React Native app.
- On Cloud PyTorch model that can verify and/or update the food image’s detected class. This powerful combination can acquire more data and process in to improve model overtime. Once food class is detected, our Graph Database of 4000 Indian Foods takes over and starts predicting the calories and estimates the best food dish that can be substituted using our algorithm. All of these interactions occur via GraphQL making the entire experience seamless and fast.
Challenges I ran into
The major challenges we faced were: a) An ML Image Recognition system that outperforms all other classifiers and can operate on low powered devices in realtime. b) An AI Recommendation system that learns and adapts to your lifestyle and diet system using graph technology. c) Integrating Graph Database and Algorithms with traditional mobile systems without losing the power and richness of graphs and relationship between data.
With resilient efforts and team work, we were able to accomplish these hurdles and provide an unique experience like no other.