Hello, I’m Ashish 👋

I’m currently a Data Scientist at Flipkart, where I work on large-scale recommendation and user representation systems. My work focuses on architecting and productionizing models inspired by cutting-edge research—such as PinnerFormer- and TransAct-style architectures—and integrating them with robust, distributed data and training pipelines. I spend a significant amount of time on contrastive learning setups (sampled softmax with logQ correction), event-based user modeling, and scaling training efficiently using DDP and multi-GPU feature-gathering strategies.

Before Flipkart, I was a Machine Learning Engineer III at HP Labs, where I designed and deployed deep learning models for HP’s Device-as-a-Service (DaaS) platform. There, I worked on user segmentation and recommendation systems for 20M+ users using LSTMs and self-attention networks, and explored evidential deep learning combined with SHAP to improve interpretability and robustness in production models.

I hold a Master of Science in Data Science from Chennai Mathematical Institute, and a bachelor’s degree in Statistics from Banaras Hindu University. My academic background strongly influences how I approach problems—I’m drawn to data-centric thinking, statistical learning, and principled modeling over ad-hoc solutions.

This blog is my personal knowledge log. I use it to document what I learn while working on machine learning, deep learning, recommendation systems, and statistical modeling—often diving into the why behind methods, not just the how. Writing helps me clarify my thinking, and sharing it is my way of giving back to the community that has shaped my journey.

If you’re interested in recommender systems, representation learning, or the practical challenges of deploying ML at scale, you’ll likely feel at home here.