Balancing mobile engineering with machine learning
Melody Yang offers insights as a mobile engineer with machine learning expertise–a rare combo in app development
Mobile engineering is a lucrative career for those who put the time into acquiring the skills and certifications needed to make it out there. It becomes a whole different ball game when you add machine learning into the mix, which is its own vertical in its own right, with its own set of unique skill sets.
What do you get when you combine the two? Melody Yang—a rising star in the mobile development space for her wide range of accomplishments stemming from her skill sets.
The fruits of being a savant
Yang recently received press for going from being an F1 visa holder to a software engineer at Apple in the past year. She additionally founded and developed six apps, two of which utilize machine learning in unconventional ways:
- Nukon is a Japanese language learning app. This app made machine learning on iOS accessible. The underlying machine learning models in Nukon were open-sourced to allow regular iOS developers to integrate Japanese handwriting recognizers into their apps in minutes. Developers can also swap out the datasets and use Yang’s training techniques to create machine learning models for other languages.
- Recogmize is an image identifying app which showcases several different machine learning models/techniques for image recognition. All the machine learning models achieve accuracies of 95+ percent.
Yang has become an expert in juggling multiple components in each of her companies by keeping priorities aligned while making use of OKR’s. It is worth noting that she developed her apps from the ground up to keep chaos to a minimum, both in the backend and for the user.
Not only “mobile first,” also the “customer first”
Despite the fact that GPT3 and other recent developments in the machine learning space are all cloud-based, Yang chose to implement her solutions locally. Her reasoning was simple—to provide the best user experience possible. While cloud-based comes with the flexibility to update machine learning models anytime, the need for network access is its downside.
Having machine learning implementations locally, namely, on-device machine learning, has better performance as it’s not relying on network requests. “Taking my app Nukon that helps people learn the Japanese language with machine learning as an example, most users build their habits of frequenting the app,” says Yang.
“It’s frustrating when the app can’t give you feedback on your learning simply because your network is out and the machine learning feature can’t function properly. Hence the users’ learning experience gets disrupted. It becomes hard for users to maintain their habits of learning. I prioritize the user experience first and foremost.”
The latest trend in app development: online learning
When asked where the future of app development is headed (with machine learning in mind,) Yang made it clear that online learning apps will reign supreme. Of course, user experience will be at the forefront. “Online learning has become a new norm,” says Yang. “More developers are focusing on apps that teach students how to learn math, languages, and more with interactive content.”
Of course, there are more components that stem from the online learning experience, especially when factoring instructors. While some teachers conduct live classes via video conferencing, there are others that choose to record videos for students to learn at their own pace at home.
While the latter may initially seem considerate, one of the frustrating points of self-paced learning is not being able to ask questions or receive feedback on demand. “This is why when I developed Nukon, I deployed custom machine learning models to give users real-time feedback on their performance in speaking/writing Japanese,” says Yang.
Yang also states that companies have developed chatbots to enable real-time question-answering on mobile devices, with the added boost of machine learning.
“These bots learn the students’ learning style and adjust to them. They are powered by machine learning models on mobile devices. There’s undoubtedly an increasing demand for the need of applying techniques in apps easily without in-depth knowledge of machine learning.”
Giving back to the developer community
Large companies such as Google also make their sophisticated, well-trained machine learning models accessible to their developer community, and Yang happily follows suit. As an iOS developer with a background in machine learning, Yang open-sourced her machine learning models in the hopes of making it easier for fellow developers that are looking to build on it and create their own mobile solutions with the added power and potential machine learning.