Krishna is a Senior Data Scientist passionate about analytics and Machine Learning to solve challenging problems in eCommerce, marketing, and NLP. With 4 years of experience in Machine Learning, including his tenure in the analytics division of JP Morgan Chase & Co., he has developed and deployed data-driven solutions for various clients across different domains and industries, building NLP pipelines in Python using Keras to extract and summarize opinions and feedback from customer product reviews, performing market mix modeling, demand forecasting, and A/B testing analytics to optimize spending, supply chain, and marketing funnel, and building data pipelines and dashboards in AWS, GCP, and Tableau to provide actionable insights and data-driven decision-making. Krishna holds certifications in Machine Learning, Python, and software testing and is keen on using data and ML to create value and impact for businesses and society.
AI, ML & LLM
Machine Learning
Deep Neural Networks
Deep Learning
LLM Agents
Large Language Models (LLMs)
LLM
Generative AI
Generative Artificial Intelligence (GenAI)
Designed, developed, and deployed a temporal GNN (Graph-LSTM, contrastive learning) for creating dynamic user embeddings, enhancing downstream ML model performance, and eliminating feature engineering time.
Developed and deployed reusable SageMaker model explainability reporting pipelines (SHAP, PDP), standardizing validation processes and reducing model debugging time by 60%, increasing stakeholder trust.
Designed a modular, reusable self-supervised learning pipeline (self-distillation), improving user tagging performance in cold-start settings.
Built customized on-disk graph back end using custom Feature Store, GraphStore implementation in PyG to scale the GNN training on huge data (500M edges & 50M nodes).
Mentored data scientists through pair-programming, peer reviews, and hiring participation.
A fully convolutional network (FCN-32s) trained to semantically segment forest scene images with RGB and nir_color input images. The project was developed to help unmanned drones in smooth navigation. The model is trained and tested on still images of forest scenes. Used Intel Edison and Microsoft Kinect for POC and prototype creation.
Worked on a smart medical network for Intel ESDC 2016, Shanghai. The project aimed to create an ecosystem of a medical network that stores patients' clinical and real-time data for smoother and quicker diagnosis in an emergency.
