• Overhauled and optimized an existing image clustering pipeline that reduced the processing time by 84%.• Architectured a pipeline using SOTA models to increase the number of objects detected in an object by 60 times.• Deployed ML Pipelines on (on‑prem) K8s and AWS Batch using Metaflow.• Contributed to developing APIs for deployment of the pipeline• Optimized CICD workflows for machine learning modules with a mix of self‑hosted and github action runners• Spearheaded in… Show more • Overhauled and optimized an existing image clustering pipeline that reduced the processing time by 84%.• Architectured a pipeline using SOTA models to increase the number of objects detected in an object by 60 times.• Deployed ML Pipelines on (on‑prem) K8s and AWS Batch using Metaflow.• Contributed to developing APIs for deployment of the pipeline• Optimized CICD workflows for machine learning modules with a mix of self‑hosted and github action runners• Spearheaded in standardizing the Logging system.• Modularized the prevailing video processing pipeline with a bespoke architecture with 60% faster processing Show less

• Revamped UI design for UniArk (Deployed at: https://uniark.org/) which 93% of stakeholders preferred.

• Authored and published 2 research papers in reputed journals like Springer IJFS and TSP CMC (First Author).• Guided a team of skilled individuals in research projects and published 1 paper in field of Computer Vision.

• Designed a novel human in the loop learning algorithm that helped in reducing the human biases in datasets that had an improvement of 5%on the benchmark datasets.• Spearheaded and collaborated 5 machine learning projects for stakeholders that revolved around reinforcement and supervised learning.

• Developed a Ticket grading feature for the clients to judge the authenticity of a false positive.• Implemented bespoke models or pipelines for different use cases using Pytorch, Tensorflow and ONNX over Triton servers with average modelprocessing time of 120 milliseconds.

• Conducted research on use of AI to improve the success rate of embryo transfer in IVF for humans (15% improvement in accuracy).• Created a synthetic image generation application using StyleGAN to enhance the existing dataset (10% difference in accuracies).• Developed a Heat map generation model for inferences learnt by the model using Grad CAM (63% of stakeholders preferred the product withthis feature).