Medic for Apache Spark - First Aid for Failing Jobs - Drasko Profirovic, Pinterest
AI Engineer World's Fair 2026 · 11:21
← All organizationsPinterest
www.pinterest.com ↗
Visual search, discovery and advertising
Pinterest is a visual search and discovery platform where people find ideas, plan projects and shop. Users save visual posts called Pins into boards, assemble collages and explore ideas for their homes, wardrobes and everyday lives. Visual search helps people find styles they cannot easily describe in words, while AI-powered boards tailor inspiration to their tastes. Businesses can create advertising accounts and use marketing tools to reach people exploring potential purchases.
Founded by Ben Silbermann, Evan Sharp and Paul Sciarra, Pinterest is now led by CEO Bill Ready. Its engineering contributions include PinSage, a recommendation system described in 2018 that combines random walks with graph convolutions to represent items using both their features and their relationships. Pinterest deployed the system on a graph containing three billion Pin and board nodes, demonstrating how graph-based learning could support recommendations across a large visual collection.
Pinterest earns revenue by delivering advertisements on its website and mobile app. The company reported 640 million monthly active users as of June 2026. Its acquisition of tvScientific, completed in February 2026 for $465.1 million in total purchase consideration, primarily cash, added a connected TV performance advertising platform, expanding its advertising business beyond its website and mobile app.
2 talks
Newest first
What We Learned from Using LLMs in Pinterest
AI Engineer World's Fair 2025 · 18:13
3 speakers at AIE
Affiliations reflect their AIE appearances, not necessarily current employment.
Start here
- What We Learned from Using LLMs in Pinterest
Start here to learn how query-Pin relevance classification combines language models, visual captions, and user signals, with reported comparisons against multilingual BERT and SearchSage.
Mukuntha Narayanan · Han WangAI Engineer World's Fair 2025
- Medic for Apache Spark - First Aid for Failing Jobs - Drasko Profirovic, Pinterest
Read this for the architectural progression from MCP-enabled tools and a single ReAct agent to a harness built with LangGraph and Deep Agents.
Drasko ProfirovicAI Engineer World's Fair 2026
Messages from the stage
Making multilingual relevance practical to serve
The search session describes distilling a multilingual teacher into a production student, selectively refreshing embeddings, and reusing relevance-tuned representations across languages and downstream tasks.
Keeping diagnostic evidence focused
Medic combines targeted exception retrieval with isolated multimodal analysis of time-series metrics. Bounded-token visual summaries help structure the evidence within an observable, regression-tested multi-agent harness.
Affiliations reflect each recorded session, not necessarily current employment.