Media performance data platform

Summary

Streaming analytics and forecasting for an advertising group.

The client needed to combine campaign data streams, monitor performance and support planning with forecasts.

A distributed platform for ingestion, storage, campaign dashboards and metric forecasting across concurrent advertising campaigns.

Tech Challenge

• The architecture was designed to scale across large campaign-data streams, including stateful and stateless events.
• Processing of huge stream of data with tf.Datasets for further pushing it to the model which is responsible for metrics forecasting.
• Configuration tf.FIFOQueue to efficiently pass train and test data into the model.
• Model optimization with TensorRT which doesn't support numerous operations.

Solution

• Our team has created a whole stack of technologies and corresponding dashboards to visualize the analyzed data and current campaigns performance.
• Platform was built as a scalable micro-services solution with the ability to process data from hundreds of campaigns successively, asynchronously and in parallel.
• Additional work was done to handle stateless and stateful data points in streams.

Impact

A media analytics platform combining data processing, dashboards and forecasting.