The applied science team works on frontier recommendation systems, user representations, and LLMs at Pinterest. This team fully leverages the unique graph dataset from boards, pins, search queries, videos, collages, and more, to build the systems that deeply understand the stylistic and aesthetic trends that underpin engagement on the entire platform. Applied science regularly publishes its work on large-scale graph learning and long-sequence user modeling.

The Visual team in labs focuses on all aspects of visual understanding and visual generative modeling. The team has its origins in large-scale visual representation learning and other core visual signals, and in recent years has expanded to multimodal pre-training for all content-level understanding on the platform. The team also builds Pinterest Canvas, a diffusion-based visual generative model trained from scratch on the text-image-video content graph, and fine-tuned with visual instruction datasets to enable open-ended image editing and enchancement.

The Responsible AI team pioneers innovative ML solutions to ensure Pinterest remains a positive, inclusive and fair platform for all users. Working in the gray area beyond traditional policy enforcement and rooted in solving real user problems, this is the team that develops the models behind Pinterest's skin tone, hair pattern, and body type search refinements. They also work in collaboration with partner teams across Pinterest to evaluate ML systems for fairness, and create and integrate fairness-aware ranking algorithms to diversify content recommendations where needed. Lastly, this team has spearheaded Pinterest's comprehensive framework for responsible GenAI product development, including evaluation methodologies, red teaming approaches, and mitigation strategies.

The AI Products team at Pinterest Labs accelerates cutting-edge technology from research to real-world impact. We are a multidisciplinary group spanning product engineering, design, content, data science, and product management, dedicated to rapidly prototyping and incubating next-generation AI and machine learning products. By embedding a full-stack team with Labs’ ML researchers and engineers, we are uniquely positioned to pioneer innovative, user-facing capabilities, especially in generative AI, where breakthroughs demand entirely new user experiences.

The Performance & Prototyping (P&P) team acts as the infrastructure innovation team for Labs. Many of the newest modeling breakthroughs across RecSys, LLMs, diffusion models, and new multimodal representation models require experimental ML infrastructure that first needs to be proven out in a fast-moving setting before being absorbed into the larger ML Platform team at Pinterest. P&P thus partners directly with ML researchers at Applied Science and Visual to co-develop training pipelines, kernel optimizations, graph model samplers, inference infrastructure, and more, as new techniques emerge.