At Amazon, I built an LLM-powered Ads Copilot that reached 25,000 weekly active users in its first six months, and developed fine-tuned models for intent understanding, ad objective recommendations, and multilingual content moderation. My work improved moderation automation by 15.8%, enabled auto-approval of 500K+ answers, and saved $100K in labeling costs.
I've also built RAG plagiarism detection, multilingual question-answering systems, and ML models for ticket classification and course-dropout prediction. I work across Python, PyTorch, Hugging Face, LangChain, AWS, and GCP, with hands-on experience in SFT, DPO, LoRA, quantization, retrieval, and model evaluation.
