Contributed to India’s pioneering L3-L5 video perception team for ADAS, focusing on enhancing object detection performance for rare classes. Developed algorithms, pipelines, and a web application to achieve this goal.Streamlined deployment pipeline for large-scale vision models as RESTful API endpoints, reducing integration time by 5x. Implemented GPU sharing amongst multiple models, reducing application costs by 75%Provisioned an easily manageable cloud infrastructure(MS Azure) utilizing IaC(Infrastructure as Code) tools like Terraform.Led development of Reranking ML models for object detection search, boosting precision by 10% without compromising recall.Directed development of compression techniques and automated deployment for large-scale vision foundation models on edge devices, optimizing performance.

Engineered an end-to-end MLOps system for a Web Application for object search using image/text prompts, covering cloud infrastructure setup, model deployment pipelines, and a custom workflow orchestrator for ML workflows.Designed an overall service architecture and workflows of the Machine Learning based object search web application with micro-services architecture.Built fully automated CI/CD pipelines on Github Actions and Azure Pipelines for containerized applications using Docker.Deployed a centralized logging and monitoring environment (Prometheus, Grafana) which captured system and model metrics, and enabled auto-scaling of model endpoints.Created a web application to analyze and score resumes using GPT-3.5 Turbo, Langchain, and Streamlit, cutting screening time by 50%.Directed a summer internship program focusing on domain adaptation of datasets through Cycle GANs and Contrastive Learning.

Automated revenue tracking attributed from digital channels, enabling instantaneous revenue data availability decreasing turnaround time by 8x for all marketing and design campaigns dollar value assessments, resulting in faster performance tracking and decision makingDesigned and delivered comprehensive dashboards, marketing campaigns analyses and executive scorecards tracking performance and relaying key actionable business insights for global revenue and personalized offers for all digital channels

Comprehensively studied and implemented a state-of-the-art deep recurrent visual attention model for predicting breast cancer on histopathology images as baseline.Formulated and designed a new recurrent U-Net convolution architecture for the same.