The Reco team is responsible for users experience from on-boarding to adding products in the bag. Currently working on Machine Learning models to show the most relevant products to users based on their usage pattern and buying propensities.
A strong personalisation system is very important for us given the thousands of products that we have on the platform. Each less relevant product we show the user, is an opportunity lost for us. The more relevant products we show the users, the less time and effort users will need to find the products they love, enhancing the user experience.
The team is building a large scale recommendation engine which will capture user activities and suggest products, categories, brands and even styles to the user. The beauty of this engine lies in the scale at which it can operate.
The user activities are fed to a scoring engine called Stream-Mapper. It ranks the activities based on its various attributes. The ranking gets updated based on the frequency of the occurrence of the activities. Finally the Personalization framework (Pipeline) takes up the scores for each user and calls brands, categories and products from the inventory.
To power the recommendation engine the team had to build a Reactive Grid for fast bucketing of inventory. By virtue of this, recommendation engine can generate suggestions for millions of users within a fraction of a second.
Fynd is known for its well thought out and crafted user interface on our customer facing products. But while building the recommendation engine the team totally revamped the UI of the apps in order to showcase the user recommendations. This is a huge project involving individuals from business, design and engineering team. The team has re-engineered our Android & iOS native apps (RIP Windows) and Web & mobile web to cater for personalisation.
The team, hence is tasked with building a large scale experimentation engine Galvatron aka galvy to test how the system will behave based on the user insights. This allows the Growth team to run and iterate quickly on the experiments. These experiments are based on recommendation types, visibility ranking of brands & product categories etc. The team also built a product called Vortex to manipulate and monitor these experiments. The team is also building a Real-Time In-App Notification Service called LightSpeed powered by the recommendation engine.