Applecart

We have three separate teams within the engineering organization. Two focused on a new direct-to-consumer product and one focused on a legacy B2B product. These positions are for our direct-to-consumer product, where there is a recommendation team and an app team.

The recommendation team provides real-time ML predictions around what brands a user may like and what contacts a user should refer an offer to. This team leverages Flask to serve the microservice. The app team is responsible for building the mobile application as well as a REST API that supports mobile application. The mobile app is written in React Native and the backend service is written in Django with django-rest-framework.

Our organization runs on agile with a weekly sprint cycle. We have our iteration planning meetings on Mondays and standups every other day of the week. Every engineer and data scientist works on tickets that are spec’d and estimated in the week prior. Every person is actively involved in both the spec’ing and estimation process, as we like everyone to have strong ownership over one’s work product.

We have retrospectives every two weeks to reflect on what has gone well and what could have gone better over the past releases, both specific to new features as well as general process. We pride ourselves on having a bias towards action, so we always push to create and assign an action item for the key themes of what could be going better.

As we serve real-time recommendations, we have a number of challenges with operating microservices. Namely, keeping data in sync (both in serialized data science models as well as tables on the recommendation API) and retraining and calibrating machine learning models.

We provide consumers with a way to be compensated for referring their friends to the brands and products they already love and use. Because we compensate consumers, we anticipate a level of fraudulent activity. To deal with this today, we’ve developed human-in-the-loop processes for verification but we anticipate a number of challenging technical problems in this domain.

The product is quite young, meaning that we have a number of open challenges on the systems and devOps level, such as improving our level of end-to-end test coverage across the two backend services, snapshotting and sanitizing data for development and testing, and having backward compatible API releases as submission to app stores can create a bottleneck.