Projects
A selection of projects I’ve worked on recently. More details and code can be shared upon request.
Computational Pathology
Applicable AI System
Designed and implemented a web-based viewer for gigapixel whole slide images (WSIs) that leverages an unsupervised,
pan-disease anomaly detection framework to highlight regions of potential pathological significance.
The system integrates statistical modeling with machine learning–driven feature representations,
enabling quantitative heatmaps and assist-on/off diagnostic modes.
It is currently being tested with GI pathologists to evaluate its impact on diagnostic workflow and efficiency.
AI for Health
Clinical Text
Uncertainty-aware Inference
Developed a probabilistic baseline that combines simple feature extraction with calibrated
uncertainty estimates to rival large language models on clinical reasoning benchmarks, at a
fraction of the computational cost. The method aims to be deployment-friendly for real hospitals.
Multimodal Learning
Clinical Reasoning
Representation Alignment
Developing methods to align representations across ECG signals, echocardiography videos, and CT imaging
using contrastive learning and temporal encoders. The goal is to enable cross-modal retrieval, improve
diagnostic consistency, and provide clinically grounded predictions when only partial modalities are available.
Working under the supervision of Prof. Ricardo Henao,
this project explores how shared latent spaces can support cardiopulmonary risk stratification and early-stage clinical decision support.
