Over the past 12 months, Ishita Prasad has led the development of foundational AI capabilities for embodied agents at Meta, advancing both core research and real-world deployment. She drove the launch of Segment Anything (SAM), a scalable, open-source framework for multimodal segmentation, which reached 100K+ GitHub stars and was adopted by 200+ research and enterprise teams worldwide.
She also spearheaded PARTNR, the largest benchmark for planning and reasoning in household robot tasks, encompassing 100K+ language tasks, 60 homes, and 5,819 objects. This work revealed key limitations in SoTA LLMs and demonstrated that smaller, fine-tuned models can match performance of models 9x larger while being 8.6x faster, shaping Meta’s roadmap for efficient embodied AI.
Additionally, she led the development of LOCATE 3D, setting new SOTA in 3D referential grounding, enabling real-time deployment in robots and AR. Beyond product and research impact, she mentored 12+ junior PMs and researchers and contributed to published AI papers.
Is this your organization?
Claim your organization profile to manage awards, description, and public presence.