Experience

Research Fellow

Microsoft Research New England -- Cambridge, MA

June 2024 — June 2026

  • Co-developed contrastive multimodal representation learning models fusing microscopy images with molecular structure data, outperforming baselines on drug-screening retrieval by 70% (PLOS Computational Biology)

  • Trained a 775M-parameter autoregressive transformer generating protein localization images from amino acid

    sequence, reducing FID from 33.0 to 6.9 vs. prior best method, with zero-shot transfer to unseen imaging setups

  • Built and open-sourced MorphoHELM (first author), a benchmark comparing 8 image representation methods

    across 5 public datasets (2,700 plates and 42K drug and gene perturbations from 11 institutions) quantifying how

    performance degrades across four levels of experimental noise

  • Designed the evaluation protocol using permutation testing, correcting two flaws in the field’s standard metrics: an

    imputation scheme that skewed enrichment scores, and clustering metrics that reflected embedding geometry rather

    than biological signal; showed general-purpose vision models outperform every microscopy-specific model tested

  • Scaled pipelines processing 10M+ microscopy images from 12 data-generating institutions (download, illumination

    correction, quality control, format conversion) packaged as internal libraries and HuggingFace datasets

  • Ran distributed training and inference on Azure across clusters of up to 8 H100 GPUs, using PyTorch DDP and

    FSDP, Docker-containerized environments, and Weights & Biases experiment tracking

Undergraduate Research Assistant

University of Michigan -- Ann Arbor, MI

May 2023 — May 2024

  • Designed a PyTorch 3D variational autoencoder predicting medication effectiveness for chronic pain patients from fMRI scans, pretrained unsupervised on 1,000 patients before fine-tuning on labeled pain scores

  • Built a Boto3 pipeline to download, normalize, and quality-filter a 1.5TB Human Connectome Project fMRI dataset from AWS S3; validated predictions with physicians at the U-M School of Medicine

Teaching Assistant for EECS445: Machine Learning

  • Facilitated weekly discussion sessions for 25+ undergraduate students on fundamental machine learning methods

  • Composed and verified educational resources, including projects, homework submissions, and presentation slides, to maintain a high standard of education

  • Oversaw the management of missed assignments for a large student body of 300+ individuals, prioritizing individual support to maintain high student morale and academic success

University of Michigan -- Ann Arbor, MI

Jan. 2023 — May 2024

Education

University of Michigan, Bachelor of Science in Computer Science

Aug. 2020 — May 2024

Honors: summa cum laude