DE/DS/ML PROJECTS

Conducted an in-depth analysis of patient enrollment trends, utilizing two distinct sources to forecast future growth. The results revealed a significant discrepancy between the projected numbers from each source, indicating potential errors or limitations in the underlying assumptions. Our investigation identified key factors contributing to this disparity, including differences in data sources and methodologies. By examining these discrepancies, we gained valuable insights into the complexities of patient enrollment patterns and can refine our models to provide more accurate predictions. This analysis enables informed decision-making for healthcare organizations seeking to optimize their patient acquisition strategies.

Developed an R Shiny application that enables users to standardize and simplify both HCO (Healthcare Organizations) and HCP (Healthcare Providers) datasets. The app accepts raw xlsx files from H1, allowing users to easily import and manipulate their data. Our solution provides a range of features, including tiered ranking columns, to augment the datasets with additional insights. Users can select from various standardization options, such as normalization or aggregation, to simplify their data and gain deeper understanding of their healthcare claims. The app's intuitive interface ensures seamless data analysis and reporting.

A World for Every Child aims to revolutionize childhood cancer diagnosis in under-resourced areas by harnessing advanced machine learning technologies. Our mission is to improve early and accurate detection of Acute Lymphoblastic Leukemia from blood smears, a critical step towards effective treatment. Despite the challenges of deploying this technology in developing nations, including resource limitations and FDA clearance requirements, our tool offers a quick, cost-effective preliminary diagnosis to aid doctors. By bridging the gap in global childhood cancer outcomes, we aim to introduce machine learning techniques to areas facing significant disparities in medical care, ultimately saving lives.

Developed a machine learning model utilizing the EfficientNetB7 architecture to classify Acute Lymphoblastic Leukemia (ALL) from blood smear images. To enhance model performance, we applied image augmentation techniques to our training dataset, increasing its size and diversity. This involved applying random rotations, flips, and color jittering to the original images. The augmented dataset improved the model's accuracy by ~13%, achieving an overall classification rate of 98.9%. Our solution enables rapid diagnosis and treatment planning for patients with ALL, reducing manual analysis time and improving patient outcomes.

Developed a Named Entity Recognition (NER) model using the Spacy Python package to augment feature engineering for identifying customer-relevant data points. The model accurately identified entities such as names, locations, and organizations from unstructured text data, including customer feedback and reviews. By leveraging this NER model, we were able to extract valuable insights into customer behavior, preferences, and demographics, enabling more targeted marketing efforts and improved customer engagement. The model's accuracy was evaluated using standard metrics, resulting in a high precision rate of 95% and recall rate of 90%. This enhanced feature engineering capabilities.

Designed a SnowPark Python-based data pipeline to process policy disbursement data from various sources, including agent and policyholder information. The pipeline integrates with Snowflake's scalable architecture to feed real-time data into weekly Qualtrics surveys via custom APIs. Additionally, the pipeline powers R Shiny applications for real-time reporting, enabling agents and stakeholders to access up-to-date insights on policy performance and customer satisfaction. Our solution utilizes SnowPark Python's advanced data processing capabilities to transform and load data into Snowflake, ensuring seamless integration with existing systems and providing actionable intelligence for business decision-making.

Developed a custom Python wrapper package that leverages the Polars library to securely serve encrypted database service account credentials. The package utilizes encryption algorithms to protect sensitive information, ensuring seamless integration with existing data pipelines and workflows. Our solution provides a streamlined development experience for designated data science team members, enabling easy access to required credentials without compromising security. By automating credential management, we enhance usability and efficiency, reducing manual effort and improving overall productivity. This custom package ensures secure and reliable access to sensitive database information, supporting informed decision-making in data-driven applications.

Developed a custom SnowPark Python script that accurately identifies callers as either agents or policyholders based on Genesys IVR and Tax Identification Number (TIN) data. The script leverages machine learning algorithms to analyze caller interactions with the IVR system, allowing for precise classification of callers into distinct segments. Our solution enables internal product teams to target specific firms with tailored use of IVR for commonly mistaken requests, improving overall customer experience and reducing unnecessary calls. By automating this process, we enhance efficiency and accuracy in handling customer inquiries, ensuring seamless interactions between agents and policyholders.

Developed predictive modeling solutions that integrated multiple NLP techniques, including topic modeling, sentiment analysis, and keyword extraction. Combined these models to create a comprehensive understanding of customer interactions and preferences. Utilized TensorFlow ANN and LSTM layer to process large volumes of Nexidia and Genesys-derived transcripts, improving resource allocation by identifying high-priority calls and streamlining agent workflows. The solution also enhanced customer satisfaction through personalized responses and proactive issue resolution, resulting in increased efficiency and reduced wait times for customers. This pipeline improved overall call center operations, leading to enhanced customer experience and business outcomes.

Developed a SnowPark Python-based data pipeline to categorize insurance agents into three categories: Active, Inactive, and In Danger. The pipeline processes large datasets of agent information, including policy assignments, claims history, and performance metrics. Using Snowflake's scalable architecture, we designed a data flow that integrates with various data sources, including CRM systems and claims databases. Our solution utilizes SnowPark Python's advanced data processing capabilities to identify patterns and anomalies in the data, enabling accurate categorization of agents based on their performance and risk profile. This pipeline improves agent management and reduces churn.

Developed an R Plumber API model and custom R package, customeRinsight, to securely serve encrypted database service account credentials to designated data science team members. The package utilizes the encryptr library to encrypt credentials, which are then stored in a secure environment variable. Our solution provides a streamlined development experience by allowing team members to easily access required credentials without compromising security. The API model is built using Plumber, enabling seamless integration with existing R workflows and data pipelines. This custom package enhances usability and security for our data science team, ensuring efficient collaboration and data-driven insights.

A custom Selenium script is built using Python to automate routine tasks on the Qualtrics survey platform. This script leverages Selenium's capabilities to interact with web pages, filling out forms and submitting responses with ease. The script automates tasks such as data entry, form validation, and submission, freeing up time for more strategic activities. With its ability to handle complex interactions and adapt to changing page layouts, the script ensures seamless automation of routine tasks. By integrating this script into the Qualtrics platform, users can streamline their workflow, reduce manual labor, and increase overall efficiency.

Customer churn is a significant concern in annuity insurance, leading to revenue losses and reputation damage. Our team developed a machine learning model that predicts customer churn based on policyholder behavior, demographic data, and claim history. We collected ~100,000 customers' data and used Random Forest, Gradient Boosting, and Neural Networks to develop a predictive model. The best-performing model achieved 92% accuracy and 0.85 F1-score, indicating high confidence in predicting customer churn. Demographic factors like age, income, and occupation were significant predictors. The model identified key risk factors for targeted retention efforts, demonstrating the effectiveness of predictive modeling in annuity insurance.

Designed and developed an annual Relationship NPS Survey in Qualtrics, utilizing custom JavaScript and jQuery to create a seamless user experience. The survey collects critical feedback on customer relationships, satisfaction, and loyalty, providing actionable insights for business improvement. By leveraging custom coding, we optimized survey performance, reduced response times, and improved overall user engagement. The resulting NPS scores enable informed decision-making, driving growth and revenue through enhanced customer experiences and targeted relationship-building strategies. Regular surveys ensure continuous improvement.

This study employed machine learning techniques to perform causal inference in annuity insurance. A dataset of ~1,000 customers was analyzed to identify relationships between policyholder behavior and outcomes. The LATE (Local Average Treatment Effect) method was used to estimate treatment effects, revealing significant causal relationships between premium payments and claim frequency. Policyholders with higher premium payments were more likely to file claims, while those with lower payments were less likely to do so. This analysis provides insights for insurers to improve risk assessment and underwriting decisions, enabling targeted retention strategies to reduce churn and increase revenue. Accurate causal inference is crucial for informed decision-making.

Developed an R Shiny dashboard deployed in Posit Connect, providing actionable insights into customer segmentation and behavior. The dashboard combines transactional NPS survey responses with internal sales data to generate detailed segmentation reports. Users can explore trends and patterns across various segments, including cycle time comparison for first-timer vs seasoned, products, and engagement metrics. Our solution leverages machine learning algorithms to identify high-value segments and provide recommendations for targeted marketing campaigns. The dashboard's intuitive interface enables stakeholders to easily navigate and visualize complex data, driving informed decision-making and improved customer engagement strategies.

Developed a personal machine learning project to train CNN models using Nvidia Tesla K80 and RTX 2070 Super GPUs, focusing on classifying real-world cell images as parasitized or uninfected. The dataset consisted of high-resolution images from various sources, including scientific literature and online repositories. By leveraging the power of deep learning, I trained and fine-tuned multiple CNN architectures to achieve accurate classification rates. The model demonstrated impressive performance, achieving an accuracy rate of 95% on the test set. This project showcases my ability to apply machine learning techniques to real-world problems, with potential applications in medical diagnosis and research.

Created an R Markdown document deployed in Posit Connect, providing actionable insights into the cycle time between each process for new policy sales. The document calculates the time taken for each stage of the sales process, from initial application to final policy issuance. Additionally, we analyzed Not In Good Order (NIGO) items across all stages, identifying trends and patterns that inform monthly performance metrics. Our solution enables stakeholders to track key performance indicators (KPIs), such as average cycle time and NIGO rates, to optimize the sales process and improve overall efficiency.

Designed an R-based data pipeline deployed via Posit Connect, feeding new policy transaction data from various sources to support agent performance analysis. The pipeline processes and transforms data on a scheduled monthly basis, enabling timely insights into agent behavior and policy performance. Our solution integrates with Qualtrics surveys, which are sent monthly to agents via custom APIs. Additionally, we developed R Shiny applications for monthly reporting, providing a user-friendly interface for stakeholders to access key metrics and trends. The pipeline ensures seamless integration of data from multiple sources, enabling data-driven decision-making and improved agent performance management.

Built a new business transaction survey in Qualtrics, utilizing Custom JavaScript and jQuery to enhance the user experience. The survey collects critical data on monthly new annuity policy sales, including customer demographics, purchase behavior, and satisfaction metrics. Our solution integrates seamlessly with existing reporting systems, ensuring accurate and timely data collection. By leveraging custom coding, we optimized survey performance, reduced response times, and improved overall user engagement. The resulting insights enable informed business decisions, driving growth and revenue for our clients through enhanced policy offerings and targeted marketing strategies.

Developed a D3.js personal project focused on creating engaging data visualizations. The project aimed to showcase trends and patterns in various datasets, including economic indicators, population growth, and environmental metrics. Using the D3 library, I crafted interactive charts, maps, and graphs that provided insights into complex data sets. The project allowed me to explore creative visualization techniques, experiment with different design elements, and share knowledge with others through online platforms. Although the pandemic shifted focus, this project remains a testament to my passion for data-driven storytelling and visual communication.

Updated the Jeweler's insurance underwriting application written in VBA to accommodate the carrier change to Tokio Marine America. The new application streamlined the underwriting process, automating quote generation and integrating with a data pipeline to Azure SQL database for reporting. Our solution reduced manual effort by 30%, increasing efficiency and accuracy. The updated app now leverages advanced analytics to identify high-risk policies, enabling informed decision-making and improved risk management. By adapting to the carrier change, we ensured seamless integration with Tokio Marine America's systems, enhancing our clients' underwriting experience and driving business growth.

Conducted an annual reinsurance analysis for Jeweler's insurance, providing critical insights to inform risk management decisions. Our solution analyzed historical claims data, market trends, and industry benchmarks to assess the company's reinsurance needs. The analysis identified areas of high exposure and recommended targeted reinsurance strategies to mitigate risks. By leveraging advanced analytics and machine learning algorithms, we optimized reinsurance coverage, reducing costs by 15% and improving overall risk management. Our annual reinsurance analysis ensures Jeweler's Insurance remains adequately capitalized and resilient in the face of changing market conditions and emerging risks.

Developed a simple Python-based terminal app that enables underwriters to efficiently search for and modify insurance riders. The app provides an intuitive interface, allowing users to quickly locate specific riders within their policies. Our solution includes features such as filtering, sorting, and editing rider details. Underwriters can easily make changes to policy riders, ensuring accuracy and reducing errors. The app's simplicity and ease of use enhance the underwriting experience, enabling faster and more accurate policy modifications. This tool streamlines the process, improving productivity and customer satisfaction for insurance companies.

Developed a VBA-based underwriting application for jewelry insurance covered by Chubb, streamlining the process with auto-generated quotes and integrating it with a data pipeline. The app automates the underwriting process, reducing manual effort and increasing efficiency. Our solution generates accurate quotes based on policyholder information, coverage options, and risk assessment. The data pipeline feeds into an Azure SQL database, enabling real-time reporting and analytics. This application enhances the underwriting experience for insurance agents, while also improving accuracy and reducing errors. By automating the process, we enable faster quote generation and improved customer satisfaction.

Developed a VBA-based Business Lead Tracker to simplify status updates for potential new business leads. Migrating from Salesforce, our solution streamlined lead management by consolidating multiple platforms into one intuitive interface. The tracker enabled easy tracking of lead interactions, follow-ups, and conversions, reducing manual effort and increasing efficiency. By simplifying the status update process, we improved lead engagement and conversion rates, ultimately driving revenue growth for the business. Our custom solution met the unique needs of our organization, providing a tailored platform for managing complex sales data in a user-friendly manner.