At Flipkart, I developed and productionized LLM-powered solutions for Flipkart Food and Insurance Claims, including automated customer-experience workflows. I also built a post-chat intent classifier to support structured analysis of customer conversations.
For Flipkart Food, I developed a real-time rider-chat moderation system to detect and block abusive language, PII, and unsafe content. I own ML pipelines from model development and evaluation through deployment, monitoring, and maintenance.
At Google, I engineered a Go and Kubernetes proof of concept to automate Major Version Upgrades for AlloyDB, replacing a 10-step manual process. At Joly AI, I developed a PyTorch and OpenCV model that identified target facial expressions with 70% accuracy.
My research and projects include sound-source localization, reinforcement learning, and language models. I trained a 150M-parameter multilingual transformer from scratch for English, Maithili, and Telugu, and developed a pipeline to analyze and mitigate gender-bias propagation in transformer models.
