Mukund Narasimhan, Engineer, Pinterest at MLconf Seattle 2017
summary
Pinterest models user interest at web scale, combining billions of Pins and 100 million Users and Boards to power recommendations. The team fuses unstructured sources (images, videos and text) with structured signals — user curation, partner content, and engagement — to produce fresh, diverse, personalized ideas. Progress includes deployed content models and end-to-end recommendation pipelines that operate at Pinterest scale, with quantitative improvements to relevance and diversity reported across feeds and discovery surfaces.
AI-written summary · about
chapters (8)
- 0:00Talk introduction
- 1:11Pinterest software evolution
- 4:27Infrastructure challenges and solutions
- 9:08Technical teams in complex projects
- 13:39Innovation in changing environments
- 17:38Key projects and platform impact
- 20:21Containers and Pinterest's future
- 21:15Cierre y reflexiones finales
related talks
32mMoving the boundaries of possible at Pinterest
21mShopping Recommendations
21mUnleashing the Power of Data Science
45mMigrating Pinterest profiles to React
19mPersonalized Content Blending In The Pinterest Homefeed
39mMachine Learning for Relevance and Serendipity
25mMigrating Pinterest profiles to React
26mEwa Dominowska, Engineering Manager, Facebook @ MLconf SEA