Unlocking audience insights by leveraging viewer behavior for TV broadcasts.

Our backend provides detailed information about the broadcast that is selected to analyze on the website, like the show’s metadata, audience data and the show’s episode video. The episode’s audience data and the show’s metadata is provided on a daily basis to BigQuery by the company. All data is stored in BigQuery and treated daily for an optimal response time. All show’s episode videos are hosted in Google Cloud Storage and the website streams the video directly from there.

On the frontend side, we used Flutter that allowed us to create a solid project structure with a strongly typed object-oriented programming language. The main components of the applications are the video player that shows the episode streamingly from GCS, the absolute audience curve chart that plotts the audience data from BigQuery, and the segment chart that is also stored in BigQuery. This last component is extensible for future extra segmentation of the broadcast, which opens up the opportunity to client’s editors to further categorize shows and get more insights.

A primary technical requirement for the project was the containerization of the entire system to enable it to run on client's Cloud Run infrastructure. The decision to use Cloud Run was driven by the need to maintain a secure network environment limited to Google Workspace users exclusively. By containerizing the system and deploying it on Cloud Run, we ensured the confidentiality and privacy of the client's data, while also benefiting from the scalability and flexibility offered by the cloud platform.

Because of this requirement, our architecture specified different Cloud Run instances for: website hosting, backend hosting, ESPv2 proxy hosting. The ESPv2 server provides us with API management features such as authentication, monitoring, and logging. This means that the backend is only accessible by the ESPv2 server and all external communications are coming to the defined endpoints in the proxy server, so that authentication tokens can be validated and only valid tokens get a response from the backend.

The login to the website is allowed to only Google Workspaces accounts, more specifically client’s own accounts and its implementation is done with Google Identity Services for Web. This sign in process will provide the necessary tokens to make a valid call to the episode services in the backend so that the frontend can receive all the information involving the episode to analyze.