At Microsoft, I design multimodal RAG and agentic frameworks using VLMs and LLMs for technical search, reducing visual-textual hallucinations by 15% and inference costs by 23%. I also build scalable LLM-driven application search pipelines across categories, stores, and regions.
Previously at Amazon, I fine-tuned LLMs for product attribute classification, improving catalog quality and reducing product returns by 7%, and deployed InsightNet, an EMNLP 2023 weakly supervised NLP framework. At Pegasystems, I built NER, document recommendation, cross-lingual NLP, intent detection, and topic classification systems for CRM and conversational AI.
