Projects
List of cool projects
Facebook Video Recommendations Quality Ranking
Meta (formerly Facebook), Menlo Park, USA
- Led Facebook Video Recommendations Quality & Integrity Ranking team, driving -XX% policy-violating video views and incorporation of See More / See Less feature in ranking for personalized content curation, resulting in -X% video hides and +Y% global video watch time.
- Spearheaded initiatives in bias reduction, watch and engagement bait mitigation and integrity improvements leveraging innovative approaches including creator and content graph expansion based detection and enforcement, deeper engagement signal learning and advanced large-scale user feedback modeling.
- Pioneered LLM-powered comment sentiment modeling for video recommendations, achieving +XX% sentiment detection accuracy and expanding multilingual capabilities, leading to -X% negative user feedback.
Facebook Video Recommendations Ranking
Meta (formerly Facebook), Menlo Park, USA
- Led Facebook Watch Core Ranking team, transforming it into Facebook’s primary video discovery platform with 1.25+ billion monthly active users and 400+ million daily watch hours. Drove +X% global Facebook daily active users and +$Y billion dollars incremental annual revenue.
- Pioneered Watch Topics/Interests Ranking initiative, implementing a two-stage multi-task deep neural network ranking system that increased global Facebook video watch time significantly. Led hierarchical granular interest tagging effort for improved content personalization across Facebook’s recommendation systems.
Facebook NewsFeed Recommendations Ranking
Meta (formerly Facebook), Menlo Park, USA
- Pioneered Facebook’s user interest topics based Explore Feed, shifting content delivery paradigm for billions of users. Patented neural embeddings technology, driving significant +X% video watch time and +Y million user connections per month across Facebook’s (then) 1.2+ billion daily active users.
- Drove modernization of recommendation models, transitioning from GBDTs to multi-task sparse neural networks. Engineered feature pipelines, model architectures and low-latency prediction systems, laying groundwork for today’s large-scale personalization models serving 2+ billion users across Facebook and Instagram.
Advanced Driver Assistance System (ADAS) Algorithms
Analog Devices, Bengaluru, India
- Specialized in real-time vision based Advanced Driver Assistance System (ADAS) algorithm development for lane departure warning, forward collision avoidance warning and traffic sign recognition applications on the proprietary ADSP-BF609 embedded processor.