Pinterest Labs

The home of machine learning research and AI product innovation

Pinterest’s machine learning investments span across frontier recommendation systems, computer vision foundation models, hyper-scale graph understanding, responsible AI product development, multimodal generative modeling, and more.

Representation learning

At the core of Pinterest are our recommendation systems; our latest progress here is OmniSage, a unified embedding system for representing Pinterest-native content: pins, boards, products, search queries, and more. This model represents a decade of graph machine learning advances done by Labs.

Learn more

Generative models

Pinterest develops frontier generative models for multiple modalities. Pinterest Canvas is a multimodal image & video diffusion model trained on the entirety of visual graph data of Pinterest, allowing for open-ended image editing and enhancement.

Learn more

Recommender systems

Our most recent advances include lifelong user sequence modeling that captures nuanced behavioral patterns over extended periods, foundation ranking models that generalize across diverse recommendation scenarios, and generative recommender systems that create personalized content experiences.

Learn more

Visual understanding

The Pinterest visual embedding system is trained with a large multimodal pre-training contrastive task, and fine-tuned with dozens of visual-specific and retrieval-specific datasets, enabling state-of-the-art visual search.

Learn more

Responsible AI

Responsible AI represents a critical research area focused on ensuring AI systems are developed ethically. At Pinterest, our team works in the gray area beyond policy enforcement, identifying opportunities where ML expertise ensures Pinterest remains positive, inclusive, and welcoming. Rooted in solving real user problems, we integrate responsible AI practices throughout the ML lifecycle. We focus on three major areas: Inclusive AI, ML Fairness, and Responsible GenAI Systems.

Learn more

AI and machine learning is central to Pinterest's work.

Machine learning is the driving force behind all of Pinterest's search, recommendation, and generative AI products.

Our process starts with the creation of rich content representations built on the extensive Pinterest data graph, which encompasses tens of billions of nodes. We develop both multimodal foundation models and Pinterest-specific embeddings like the Unified Visual Embedding, OmniSearchSage and PinnerSage. Subsequently, these models are integrated into a wide array of ranking and retrieval systems.

Pinterest Labs

The home of machine learning research and AI product innovation

Pinterest’s machine learning investments span across frontier recommendation systems, computer vision foundation models, hyper-scale graph understanding, responsible AI product development, multimodal generative modeling, and more.

Representation learning

Generative models

Recommender systems

Visual understanding

Responsible AI

Machine learning is the driving force behind all of Pinterest's search, recommendation, and generative AI products.

Join us and create the career you love