Bio
Devansh Tandon works on AI Research & Product at Meta, leading AI & recommendations teams. He is a founding member of a new AI research group (Meta Recommendation Systems Research) to develop LLM foundation models & recommendation systems across Meta: to power Instagram, Facebook, Ads.
Previously, Devansh led ML/AI teams at Google for 7 years, building the largest ML models across Ads, Search, Discover, YouTube, Gemini. He worked on YouTube's recommendation engine, which drives 70% of video watch time for 2 billion+ daily active users. At DeepMind, he incubated a new generative recommendation system using Gemini, and published multiple research papers.
Devansh graduated Magna Cum Laude from Yale University, with a BS in Computer Science and Economics.
Sessions (2)
Additional Speaker Details
Using publicly available information we constructed an analysis to help you get a feel for this speaker before deciding to attend their session.
His one recorded public talk (AI Engineer World's Fair 2025) is a production-grade walkthrough of rebuilding YouTube's billion-user recommendation stack on top of an LLM — Gemini adapted via SemanticID — rather than a chatbot demo, and at least one attendee publicly called it the best session of that day. He has since left Google (7 years, most recently Principal PM on YouTube recsys) and moved to Meta as a Principal Product Manager leading relevance and recommendations for Instagram Reels, so a new session would likely extend those hard-won LLM-for-feeds lessons to a second massive-scale feed.
Speaking style
Recent talks (1)
GitHub
@devanshtandon- devanshtandon/chatterview-newsfeed★ 0Demo app prototyping a personalized AI content feed, built with Vite/React/TypeScript/shadcn-ui
- devanshtandon/upsc-bench★ 0Benchmark testing whether frontier AI models can pass India's UPSC Civil Services Exam, using the real marking scheme and cutoffs
- devanshtandon/RBC-Classification-ConvNet★ 5Deep CNN classifying red blood cells into pathological groups, up to 97% accuracy on a Yale School of Medicine dataset
- devanshtandon/Micromouse-2015★ 0Yale Bulldog Bots Micromouse robotics competition project