Pavan Kalyan G

WORK HISTORY

EDUCATION

ABOUT PAVAN KALYAN G

As a Machine Learning Engineer with over 2 years of experience, I specialize in Natural Language Processing (NLP), Large Language Models (LLMs). In my current role at ServiceNow, I designed and developed an LLM-powered chatbot from scratch, reducing monthly ticket rates by 32% and saving in resolution expenses. By fine-tuning advanced models like Mistral-7B and Qwen-2.7B-MoE, I enhanced response accuracy while reducing online latency by 70% through quantization techniques. At Amazon, I spearheaded improvements in Alexa\'s content moderation system by fine-tuning a Multilingual DeBERTa model. This increased automation by 15.81%, auto-approving answers, and resulted in savings of in data labeling costs. I also extended chatbot capabilities to multiple languages using multilingual BERT, improving intent classification and user experience. In my recent projects, I have focused on enhancing multilingual capabilities and designing state-of-the-art question-answering systems. I fine-tuned the Llama-7B model using a curated multilingual dataset, achieving a 20% improvement in its performance. Additionally, I constructed a Dense Passage Retriever (DPR) pipeline, leveraging LABSE for language-agnostic retrieval and achieving a 10-point gain in recall over traditional language-specific approaches. I have developed advanced architectures for document-grounded question answering, combining XLM-RoBERTa for passage re-ranking and Multilingual BART for text generation. My expertise also includes building machine learning pipelines, such as using XGBoost classifiers for ticket categorization with a recall of 0.95, and implementing clustering techniques to detect emerging ticket patterns for proactive response. With proficiency in NLP pipelines, data augmentation, and optimization strategies, I am passionate about leveraging AI to solve complex real-world problems. Whether fine-tuning LLMs or designing robust architectures, I am committed to pushing the boundaries of what AI can achieve.