CrowdSource

Goal

Crowdsource by Google is a fun, gamified way for millions of people to help train Google’s AI and machine learning models, helping products like Google Translate, Maps, and Photos better serve users across regions and cultures. Crowdsource also offers open-source datasets for non-Google products to help improve machine learning around the world.

Currently, over three million users from 190 countries contribute to Crowdsource, but because the tool relies on engagement from as broad a user base as possible, it was important that the app expand beyond Android and web to include iOS — without overextending the team.

Why Flutter?

Crowdsource had previously used Flutter to implement a challenging “Smart Camera” feature, which runs an on-device machine learning model on a live camera. Despite the complexity of the feature, the Flutter development was a success, which gave the team a lot of confidence in the tool.

Their Solution

After the resounding success with the “Smart Camera” feature, the team at Crowdsource decided to rebuild the entire Android app in Flutter. Having previously used Flutter, there were no major challenges, and the team was surprised how quickly they were able to develop and release the new app.

Results

The recreated Android app helped Crowdsource reach considerably more users — from 73% of potential users to a whopping 96%. Currently, more than 20 million users have used the Crowdsource app to contribute data to Google with the express purpose of improving Google's AI for their language, culture or interest. Users are so passionate about the cause that they've self-organized into local communities across 200 cities in the world. Last quarter, the Crowdsource app collected its billionth label since inception.

With Flutter, the team found the process much more efficient, allowing them to cut their code size in half, reduce their technical debt, and double their development speed. Because of these improvements, Crowdsource is now in the process of rewriting the app with Flutter for Android and considering a rewrite for web.