What if a patient could describe their symptoms in their own language and receive an immediate assessment of how urgently they may need care? VitalVoice is a voice-first multilingual triage system that converts spoken symptoms into an urgency level and possible diagnoses. Designed with offline use in mind, the project explores whether AI can make preliminary emergency assessment faster and more accessible across language barriers.

It’s clear, professional, and immediately explains what the project is. MedLens screens PubMed titles and abstracts, records screening decisions, and exports screening logs to Excel.

NarConnect is a public health initiative developed during Yale Young Global Scholars (YYGS) to combat opioid overdoses in New Haven, Connecticut. Through free naloxone dispensing stations, community education, and strategic outreach, the project aims to make lifesaving resources more accessible and empower communities to respond to emergencies.

What happens when an AI model performs well overall but struggles with the cases that matter most? This project investigates that problem through coronary artery disease classification, comparing convolutional neural networks (CNNs), logistic regression, and widely accessible AI models. The focus goes beyond accuracy to examine how class imbalance, rare conditions, and dataset composition can affect how reliable medical AI actually is.

Can an ICU prediction model estimate several possible patient outcomes without contradicting itself over time? This project develops Joint-CE (Joint Competing-Event), a machine-learning model that treats major ICU outcomes as competing possibilities. Using MIMIC-IV and eICU-CRD, Joint-CE predicts what may happen next and when while maintaining one coherent probability distribution across outcomes and time.