Rupali Karthikeyan

• Designed and developed AI-driven enterprise applications, including an intelligent CRM platform to optimize sales
pipelines through workflow automation, priority-based task management, and real-time insights.
• Built and integrated an AI-powered sales chatbot to automate interactions and provide real-time assistance, improving
operational efficiency.
• Architected and contributed to Smart Ally, a multi-agent AI system, enabling intelligent automation across business
functions.
• Designed and implemented by the HR AI agent, delivering accurate, context-aware responses to employee queries on
policies, workflows, and operations.
• Worked across the end-to-end development lifecycle—from system design and AI integration to deployment and
continuous optimization—ensuring scalability, reliability, and performance.
• Developed modular and scalable architecture aligned with enterprise requirements and business objectives.
• Currently building a personalized AI tutoring system using local LLMs to enable adaptive, self-paced learning with a
focus on privacy and efficiency.

• Implemented knowledge distillation pipelines for large language models to build lightweight, high-performance AI systems
in the healthcare domain.
• Leveraged Bio BERT and Clinical BERT as teacher models to train compact student models with minimal performance loss.
• Reduced model size and inference time, improving scalability and deployment feasibility in clinical environments.
• Incorporated considerations for data privacy, security, and compliance in handling biomedical datasets.
• Contributed to developing sustainable AI solutions by optimizing computational efficiency and resource utilization.

• Applied advanced signal processing techniques including Short-Time Fourier Transform (STFT), Continuous
Wavelet Transform (CWT), and Wavelet Transform to analyze welding acoustic signals for defect detection.
• Preprocessed time-series data from microphone sensors, performing noise reduction, feature extraction, and data
transformation to enhance model input quality.
• Engineered meaningful time-frequency features, improving representation of acoustic patterns and boosting model
performance.
• Developed and optimized multiple machine learning models including Random Forest, SVM, K-NN, Logistic Regression,
and Gradient Boosting.
• Conducted comparative analysis of models and signal processing techniques to identify the most effective approach for
accurate defect classification.