Harjot Singh

• Led research and development of an AI-based emotional support companion using Retrieval-Augmented Generation (RAG), combining persona-based dialogue with ethical response filtering to assist users overcoming harmful habits. • Built an end-to-end AI application using Flask to process and store thousands of articles in MongoDB Atlas using MongoDB Watchers, and generated semantic embeddings with SpaCy for real-time document retrieval and contextual response generation that later got integrated using Azure CosmosDB. • Formulated Advisory on image object detection pipeline using SuperGLUE and advanced cropping techniques, enabling automatic detection and alignment of facial features for beauty clinic makeover previews. • Developed and optimized machine learning models by implementing advanced feature engineering, hyperparameter tuning, and model compression techniques to improve accuracy and computational efficiency. Ensured that the models met real-world constraints by balancing precision, inference speed, and scalability for deployment in production environments. • Applied model optimization strategies including quantization and compression for small language models (SLMs) to enable efficient mobile deployment, balancing accuracy with inference speed.

• Streamlined data pipelines using Databricks, PySpark, and SQL, achieving a 20% reduction in data migration time while ensuring scalability and integration efficiency. • Enhanced data-driven insights by optimizing machine learning models, including Linear Regression, Random Forest, and ARIMA, using Python. Improved data accuracy by 15% through feature engineering techniques such as scaling, encoding, and handling missing values. Applied statistical methods to deliver precise retirement forecasts and personalized client recommendations tailored to financial goals. • Automated testing workflows using Python, Selenium, and React to validate web meta tags and links, reducing manual QA efforts by 20%. • Collaborated cross-functionally with product teams to integrate predictive machine learning models for 401(k) pension estimators, utilizing employee data (income, lifestyle, retirement goals) to enhance the accuracy of benefit predictions, achieving a 10% faster system performance improvement and more reliable retirement planning insights for clients. • Designed and implemented a Retrieval-Augmented Generation (RAG) framework to enable machine learning-based contextual data retrieval for retirement benefits, leveraging SQL queries to streamline employee data preparation,

• Engineered JSON-RPC-based ETL pipelines on Databricks to integrate external APIs with backend systems, ensuring seamless data flow. This improved time-series forecasting accuracy for retirement benefit predictions by optimizing data preprocessing and pipeline efficiency. • Optimized machine learning workflows for a retirement chatbot hosted on Amazon S3, collaborating with actuaries to apply statistical methods like survival analysis, mortality tables, and Monte Carlo simulations to provide personalized retirement benefit recommendations tailored to client demographics. • Designed and implemented custom retrieval systems to enable real-time, machine learning-driven responses for retirement benefit queries. Leveraged time-series analysis techniques, including ARIMA and Exponential Smoothing, to forecast retirement balances based on historical data and financial trends. • Employed MLflow for efficient model management, reducing deployment time by 30% through improved experiment tracking, model versioning, and automated retraining workflows, enhancing the accuracy of predictive models. • Deployed machine learning models on AWS using S3, EC2, and SageMaker for scalable production systems, optimizing the inference pipeline to reduce latency by 25% while ensuring high availability. • Optimized API response times for Tableau dashboards by streamlining data queries and integrating caching mechanisms. Achieved a 30% reduction in load times, enhancing dashboard performance and ensuring timely insights for leadership decision making. enhancing dashboard performance and ensuring timely insights for leadership decision-making. enhancing dashboard performance and ensuring timely insights for leadership decision-making.

• Built and maintained using Databricks and Azure Data Factory to process and integrate large datasets for pension estimators, improving data integrity by 25% through effective handling of missing values, data normalization, and incremental data loading techniques. • Conducted exploratory data analysis (EDA) using Python (Pandas, NumPy) to analyze retirement benefits data. Applied statistical techniques, outlier detection (Z-scores, IQR), and visualizations (Matplotlib, Seaborn) to uncover trends and generate actionable insights for model development. • Developed and optimized machine learning models using ensemble methods such as Random Forests and Gradient Boosting, and applied statistical techniques like regression analysis and time-series forecasting to improve the accuracy of pension benefit predictions. • Designed interactive Tableau dashboards to visualize retirement savings trends, offering detailed insights into contribution growth, withdrawal patterns, and forecasting scenarios. These dashboards enabled leadership teams to make critical strategic decisions and identify areas of improvement. • Automated reporting workflows with Databricks and Power BI, creating real-time operational dashboards for management, reducing report generation time by 40% and providing key metrics for leadership and cross-functional teams to streamline operations.

- Led the transformation of website aesthetics and user experiences by spearheading the development of a dynamic website. - Utilized HTML, CSS, JavaScript, and Bootstrap to enhance the look and feel of the platform. - Deployed the website seamlessly across various devices using Express.js. - Integrated MySQL within the application of Node.js for capturing 5000+ users data from an event functionality within the project.