Domino has helped 8tracks: deploy real time predictive models, empowering non-technical stakeholders with self-service reporting tools, and easily scaling analysis on powerful infrastructure.
Real-time recommendations without engineering headaches
8tracks offers human-curated playlists, but also makes real-time recommendations for similar mixes, new artists, genres and DJs that a listener may like. To make these recommendations, 8tracks data scientist Dustin Stanbury develops sophisticated machine learning models in Python and scikit-learn.
Dustin and the software engineers needed an easy way to integrate the predictive models into the user-facing web application, which is written in Ruby. Rather than porting the models or engineering their own integration, Dustin publishes his models with one click using Domino’s API Endpoints functionality, which turns them into a REST API that the Ruby application calls hundreds of times each minute to make recommendations to listeners.
Domino handles high-availability, zero-downtime upgrades, security and logging so Dustin can focus on improving his models. “Domino lets us improve our recommendation models faster, without distracting our software engineers.”
Using Domino, Dustin can train and deploy his recommendation models rapidly, without requiring any help from 8tracks engineers. The end result is a better listening experience for 8tracks’ users.
Self service reporting tools
8tracks’ ad sales team and marketing teams need up-to-date reports to drive strategy and answer questions for label partners. Generating these reports used to be a manual, time-consuming process, but using Domino’s Launchers, 8tracks created self-service report-generation interfaces. Marketing and sales stakeholders can enter parameters into simple web forms, and under the hood,
Domino runs Python scripts that connect to MySQL, Postgres, and Redshift data sources to generate nicely formatted reports. Stakeholders can schedule recurring reports or generated updated reports with parameters on demand.
Using Launchers, for example, the marketing team can easily query traffic data to extract demographic info that can be used for infographics and press reports. Similarly, label partners can pull usage reports for specific catalogs or time periods to guide content-based marketing strategies, like sponsored playlists.
Critically, Domino lets 8tracks automate these tasks. For example, a monthly report that previously consumed an entire week of work is now accessible anytime, on demand.
New opportunities from scalable infrastructure
Domino lets data scientists easily run analysis on scalable cloud resources without any setup, and this has empowered 8tracks in numerous ways. “Domino makes it easy to run large analyses. It’s like having all the resources in the world on your laptop.”
For example, by spinning up IPython Notebooks on 32-core, 240GB machines, Dustin was able to train more computationally demanding models on larger sets of training data — leading to more accurate models. And by combining scalable infrastructure with Domino’s functionality for scheduled tasks, Dustin is able to schedule regular training runs on powerful hardware to automatically update his recommendation models to reflect new playlists, DJs, and user listening preferences.
Finally, 8tracks data scientists are able to run memory-intensive tasks to analyze results from massive A/B tests, tasks that would normally have been untenable without far more IT setup and configuration.
With Domino, 8tracks run their data science program with a fraction of the costs and people than would be required otherwise. Domino’s capabilities have been a huge benefit to their entire team, saving time and facilitating new insights that have helped maintain customer loyalty and increased new customer acquisition.