Azure Networking

• Designed and developed an RAG AI chatbot integrated with a MongoDB vector database and OpenAIAPIs to assist Duke students in finding courses they might be interested in taking• Deployed the backend on Heroku and frontend on Vercel through CI/CD pipeline using GitHub Actions• Conducted 10+ user interviews to inform the design and system architecture of LLM-powered Chatbot

• Engineered end-to-end data analytics pipeline using DBT, SQL, Python, and Metabase that extracted, transformed, and visualized data for ~100,000 users, producing insights for marketable success cases• Validated company’s care model by discovering a positive wellness trend in 74% of active users and finding a negative correlation between user engagement and user-reported anxiety• Performed A/B testing on users with 0 attended appointments and users with 1+ attended appointments to find that the anxiety for users in the latter group is declining at a faster rate

• Coordinated engineering and design efforts to usability test, implement, and ship guidance featuresthat improved the navigation experience on our platform in time for an influx of ~200 new users• Scoped, researched, ideated, and proposed image quality control user flow that incorporated threeautomations prioritizing efficiency and flexibility, improving platform scalability and user experience• Collaborated with scientific and engineering stakeholders to streamline the way we publish analysis notebooks such that they remain usable, accessible, and readable for external users

• Developed machine learning SVM and linear regression models in Python that correctly classified 80% of patient messages, then created Flask API endpoints to deploy and integrate with company’s EHR as a microservice• Analyzed and explored ~1 million textual datapoints utilizing NLP profiler, n-gram analysis and unsupervised learning techniques such as Latent Dirichlet Allocation to identify patterns and inform ML model parameters• Interpreted machine learning model using LIME and SHAP to identify that training data was poorly labeled, which resulted in retraining of model with cleaner data and an overall performance improvement by ~22%

• Led team of data analysts, engineers, and designers to ideate, define and prioritize features for a chatbot that achieved an average of 1,000 WAU and increased daily chat engagement by 29%• Defined product metrics and created Looker dashboard to track user engagement and product usability patterns, resulting in the collection of over 5,000 additional user datapoints per day• Shipped a feedback button feature that was adopted by 13% of users and facilitated the identification of ‘appointment scheduling’ and ‘prescription order status’ as the most popular chatbot use cases