Oct 2021 – Aug 2022 · 10 mos · Bengaluru, Karnataka, India

• Architected and Led End-to-End Feed Infrastructure: Directed the design and implementation of robust feed infrastructure, recommendation systems, feature stores, and data processing pipelines for Moj, resulting in a 15% increase in daily user engagement and a 30% reduction in latency.
• ◦ Advanced Candidate Generation Techniques: Spearheaded the development of various candidate generators using techniques such as FFM, GNN, word2vec, knowledge graphs, and sequence modeling, which collectively boosted content discovery rates by 18%.
• ◦ Innovative Ranking System Implementation: Built a sophisticated ranking engine integrating DeepFM and Wide & Deep networks to generate a final, personalized ranking of posts, driving a 22% improvement in click-through rates and overall user satisfaction
• ◦ Cross-Functional Team Leadership: Mentored and managed a multidisciplinary team of data scientists and ML engineers, driving a 40% improvement in project delivery speed and boosting innovation across the team.
• ◦ Scalable ML Platform Development: Engineered a high-performance ML platform capable of supporting real-time feed generation for 100 million daily active users, leading to a 20% uplift in feed relevance and a 15% increase in session durations.
• ◦ Comprehensive Feed Generation Strategy: Devised and executed a multi-faceted approach to feed generation—including cold start, candidate generation, diversity, and adaptability—that contributed to a 35% rise in user retention.

Dec 2018 – Oct 2021 · 2 yrs 10 mos · Bengaluru, Karnataka, India

• Plagiarized Content Identification: Visual Similarity Modeling: Built a pipeline using finetuned ResNet to extract image features, with similarity metrics and ANN algorithms for efficient duplicate detection. Reduced manual reviews by 60%, achieving 90% precision and 85% recall, thus enhancing content originality and integrity.
• ◦ Real-Time Inference Service: Developed a GPU-CPU hybrid service using TensorRT and XLA to deploy deep models (CNNs, transformers) with custom batch scheduling and asynchronous processing. Integrated TensorRT to optimize deep learning models—reducing precision where applicable—and employed XLA to compile and optimize operations on the CPU, enabling high batch processing with minimal overhead. Achieved an overall latency reduction of 2x and cost reduction of 3.5x.

Feb 2017 – Dec 2018 · 1 yr 10 mos · Bangaon Area, India

Aug 2016 – Feb 2017 · 6 mos · Bengaluru Area, India

May 2014 – Jul 2014 · 2 mos · Bengaluru Area, India

• Trajectory Estimation for Wearable Devices
• Accurately estimated trajectory using IMU sensor for object moving in three dimensions
• Used sensor fusion techniques with quaternion based Kalman filter for correcting orientation
• Built a 3D visualization tool using OpenGL to analyze the swing motion in Golf