Current Learning Goals
Clinical ML Researcher
Building multimodal and explainable machine learning systems
for medicine, with a quieter art shelf on the side.
About
I am Duc Nguyen (Duke), a machine learning research engineer at EKFZ, Germany.
My work sits between deep learning, multimodal AI, and clinical medicine — building pipelines with clearer inspection layers, multimodal models for pathology workflows, and interfaces that make complex AI systems easier to reason through.
Before joining EKFZ, I worked as a research MSc student at Chonnam National University, South Korea, working on machine learning for computational empathy.
Skills
Work
Clinical AI, pathology workflows, model research, and interface design.
-
STAMP-Multiplex
An end-to-end deep-learning pipeline for multiplex images, built to support pathology workflows from image ingestion through model prediction and analysis.
-
Pathology Agent
An intelligent WSI agent combining embedding-based retrieval with vision-language model reasoning to deliver robust and explainable region-of-interest ranking in histopathology slides.
-
STAMP
An open-source computational pathology pipeline from the KatherLab team for discovering and evaluating image-based biomarkers from gigapixel histopathology slides without pixel-level annotations. The peer-reviewed workflow helps clinical researchers and ML engineers run reproducible studies across multiple tumor types.
-
Multiple Appropriate Facial Reaction Generation
Research on generating multiple appropriate and diverse facial reactions in dyadic interactions using latent diffusion models. Work published at two international venues.
-
Nursing Simulation
A full web application leveraging ChatGPT to generate detailed patient scenarios and simulation-based training content for healthcare education, enhancing clinical learning through AI-driven interactive scenarios.
-
-
TalkingChat
A child-friendly chatbot with a changable talking face and voice, making the conversation feel personal, playful, and easy to engage with.
Study
What I'm studying now and the notes I keep returning to.
Study Notes
-
Implementation details of Megatron-LM, torchgpipe, and OSLO
Distributed training notes focused on engineering tradeoffs inside large-model systems.
-
FP8-LM: Training FP8 Large Language Models
Low-precision training notes with a practical implementation lens.
-
Horovod elastic training and fault tolerance
Resilience-focused notes for distributed training under unreliable hardware or workers.
-
All other unorganized notes
A larger archive of unfinished writeups, reading notes, and half-formed ideas.
Offline
Slower things that keep the rest of the work balanced.
Photography
A separate stream of observation and composition outside research work — black-and-white studies, street fragments, and quiet architectural frames.
Art
Sketches, studies, and visual notes kept loosely as a digital sketchbook.
Playlists
Loading tracks…
Loading tracks…