
Sarath Tharayil SreenivasanSarath Tharayil
Selected projects in data science, machine learning and NLP
Developed a novel architecture combining CNNs and Transformer networks for robust classification of AI-generated faces, achieving 83% accuracy. Implemented advanced data augmentation techniques and self-supervised pretraining using contrastive learning, improving model performance by 15% and validation accuracy by 12% on limited datasets.
Designed and launched a Retrieval-Augmented Generation (RAG) chatbot leveraging LangChain for conversational orchestration and ChromaDB for semantic vector-based document retrieval. Integrated sentiment analysis using transformer-based models to adjust response tone and content based on user emotions, increasing user engagement metrics by 18% and reducing negative feedback by 22%.
Devised an ML model to predict building-level energy consumption using historical meter readings, weather patterns, and building characteristics from a public dataset of over 1,400 buildings. Incorporated time-series features like lag metrics, moving averages, and seasonal encodings to enhance model performance by 12%.
Developed a predictive maintenance system for industrial equipment using sensor data and machine learning algorithms. Implemented real-time monitoring and anomaly detection to predict potential equipment failures.
Developed a machine learning model to predict customer churn with 92% accuracy. Implemented feature engineering and SHAP value analysis to identify key churn factors. Created an interactive dashboard for business stakeholders to monitor churn risk and take proactive measures.
