Jun 2025 – Oct 2025 · 4 mos

Apr 2023 – Jul 2025 · 2 yrs 3 mos

• Enhanced personalization by designing a deep learning-based ranking model combining DNN and Flash Attention Transformers. Resulted in 18% improvement in offline metrics and 10% boost in online engagement KPIs.
• Led multimodal innovation by developing a context-aware Video Captioning model integrating ViT, BERT, and cross-attention decoders. Used KL divergence loss for robust training and richer outputs.
• Improved user retention through dynamic trending content detection via density-based clustering and metadata analysis, improving engagement metrics by 8%.
• Optimized training and inference workflows using Ray for distributed training and Nvidia Triton for scalable serving. Enabled real-time recommendations with flexible model versioning and custom logic scripting.

Aug 2021 – Mar 2023 · 1 yr 7 mos

• Boosted video content discoverability by building a scalable face identification pipeline (batch + stream) for celebrity recognition, enhancing accuracy by 20% and slashing latency by 200% using Adaptive Marginal Guidance and BRISQUE-based quality filters.
• Drove major traffic gains by deploying Personalized Page Rank and SimClusters retrieval arms, contributing 60%+ of platform traffic and achieving 12% uplift in recall@10 and CTR
• Standardized infrastructure across cloud platforms (AWS, GCP, Azure) and orchestrated pipelines using EMR, Sagemaker, Azure ML, and BigQuery to scale ML systems across millions of users

Mar 2021 – Jul 2021 · 4 mos · India

• Streamlined cloud infrastructure tasks for Project LADE, using VMware tools like vCenter and piloting vRO to automate provisioning and monitoring.
• Delivered infrastructure pilots for high-profile automotive clients including McLaren and Audi, ensuring secure, scalable, and robust systems for their cloud environments.
• Contributed to operational efficiency by scripting workflows that enhanced visibility and control over distributed compute systems.