Duc Nghiem Do

Data Science student · Winnipeg, Canada

Relevant Experience

Institute for Minimal Access Surgical Sciences
St. Boniface General Hospital

Data Analyst (Co-op)
September 2026 – Present · Winnipeg, Manitoba

  • Conducting multimodal research involving eye-hand tracking and facial expression data from surgical simulation studies.
  • Managing and cleaning research datasets, performing exploratory analysis and literature review, maintaining the lab website, and providing computational support.

Digital Weeds Lab, University of Manitoba

Research Assistant (Co-op)
April 2026 – September 2026 · Winnipeg, Manitoba

  • Researched AI for plant science with a focus on crop and weed semantic segmentation.
  • Performed manual image segmentation in QGIS for dataset annotation, exploratory analysis, literature review, and computer vision tasks.

PASSIO Lab, North Carolina A&T State University

Research Assistant (Part-time)
October 2025 – Present · Remote

  • Researching AI in healthcare with a focus on cardiology applications.
  • Investigated ECG signal analysis through feature extraction and selection, dimensionality reduction, unsupervised clustering, and disease phenotype modeling.
  • Developed and evaluated self-supervised learning and fine-tuning approaches for ECG foundation models, using explainable AI for clinical interpretability.
  • Contributed one first-author accepted paper at CAI 2026, with further research in progress.

Database and Data Mining Lab, University of Manitoba

Research Assistant
May 2025 – October 2025 · Winnipeg, Manitoba

  • Collaborated on breast cancer detection from mammogram scans and LLM quantization for improved memory efficiency and inference performance.
  • Processed 55,000 mammogram scans from 12,000 patients with severe class imbalance, applying preprocessing, multimodal and multitask learning, and imbalance-handling strategies.
  • Compared five CNN and Transformer backbones, including ResNet-50 and ViT, with up to 86 million parameters. Ablation studies covered external datasets, modality configurations, and imbalance-handling methods, achieving 71% balanced accuracy, one percentage point above the referenced benchmark.
  • Quantized six Hugging Face LLMs using C-based scripts for weight extraction and compared dequantized weight statistics against base models.

Linagora Vietnam

Software Intern
August 2025 – September 2025 · Hanoi, Vietnam

  • Developed frontend features for Twake Calendar using TypeScript and React in an open-source codebase.
  • Collaborated across Java, JavaScript, DevOps, QA, and Mobile Flutter teams on product workflows and consistent cross-platform behaviour.