Chentao Jin

Led development of a fairness-aware skin cancer diagnosis classifier using a biased, imbalanced subset of the Fitzpatrick17K dataset. Improved AUC from 0.676 to 0.824 and reduced skin tone bias nearly to zero by upgrading a baseline CNN to a ViT-B/32 + BERT multimodal model. Generated structured lesion descriptions utilizing Gemini and MONET models, integrated various image encoders and text transformers using multiple fusion techniques, assessing performance-equity trade-offs in multimodal architectures.

Implemented Python execution in a backend server running on Docker containers by modifying Docker files and mapping volumes. Developed the Umple compiler to support external Python execution and implemented execution functions in the Umple Online platform. Built, tested using JUnit and Ruby, and deployed the program using CI/CD tools such as Apache Ant, Jenkins, and AppVeyor to accelerate the DevOps lifecycle.

Designed and built a database and data management system using PostgreSQL and MongoDB for an employee management platform. Implemented security measures (Row Level Security) for secure data access with varying user privileges, handled slowly changing data issues by adding type-2 valid-time attributes, and created periodic aggregates for query optimization. Developed and tested RESTful APIs using Node.js for microservices, implemented authentication mechanisms for secure data retrieval and reliable business operations.