Data Architect Senior

The Machine Learning in Neurosurgery (MLiNS) Lab at the University of Michigan is recruiting a Data Scientist / Machine Learning Engineer to build and lead the multimodal data infrastructure behind our next generation of medical AI systems. You will work with health-system data spanning radiology, digital pathology, intraoperative microscopy, and longitudinal electronic medical records, transforming complex clinical data into reliable, governed, and reusable scientific infrastructure.

This is a high-ownership role for an engineer-scientist who wants to work at the boundary of machine learning, clinical medicine, and large-scale biomedical data. Your work will directly enable foundation models, vision-language systems, and AI agents designed to improve diagnosis, surgical decision-making, and patient care.

Responsibilities*

• Design and maintain scalable pipelines for ingesting, harmonizing, linking, and versioning multimodal clinical data across radiology, digital pathology, intraoperative imaging, and electronic medical records.
• Develop durable SQL data models, metadata standards, cohort-building tools, and interfaces that connect imaging, pathology, clinical text, procedures, treatments, and longitudinal outcomes.
• Create robust preprocessing pipelines for DICOM studies, volumetric MRI and CT, whole-slide pathology, and stimulated Raman histology.
• Establish automated data-quality monitoring, validation, lineage, provenance, de-identification, and audit processes for HIPAA-regulated research environments.
• Support distributed model training and evaluation on high-performance computing and cloud infrastructure using reproducible environments and modern MLOps practices.
• Contribute to medical foundation models, vision-language models, clinical NLP systems, and AI agents that operate over multimodal health-system data.
• Collaborate with clinicians, scientists, trainees, and engineers, and contribute to manuscripts, datasets, open-source software, conference presentations, and high-impact publications. Opportunities exist to lead independent technical and scientific projects, * Our core research programs include intelligent histology, AI-based neuroimaging, visual intelligence, patient forecasting, and collaborative neuro-oncology. We work closely with clinicians, pathologists, radiologists, computer scientists, trainees, and research engineers to move ideas from scientific discovery to clinical evaluation. Learn more: www.mlins.org

Our Scientific Paradigm: Health System Learning

We are building a new paradigm for medical AI called health system learning. Rather than relying only on small, manually curated datasets, we develop secure and reproducible systems that learn from the multimodal data generated during routine clinical care. Our long-term aim is to enable AI agents to learn within the clinical environment, grounded in imaging, pathology, clinical text, workflows, treatments, and patient outcomes.

The person in this role will create the data substrate that enables this vision. You will help define how clinical data are organized, linked, quality-controlled, versioned, governed, and made usable for large-scale learning while maintaining rigorous standards for privacy, security, provenance, reproducibility, and scientific validity. Recent Work from the Lab

The lab has a strong record of publishing and translating high-impact medical AI research, including:

• Nature Medicine (2026): Health system learning enables generalist neuroimaging models NeuroVFM, trained on 5.24 million clinical MRI and CT volumes.
• Nature Biomedical Engineering (2026): Learning neuroimaging models from health system-scale data Prima, a foundation model evaluated in a health system-wide clinical study.
• Nature (2025): Foundation models for fast, label-free detection of glioma infiltration FastGlioma for real-time detection of tumor infiltration during surgery.
• CVPR (2026): ItemizedCLIP and CodeV New methods for complete visual representations and faithful agentic visual reasoning; CodeV was selected as an oral paper.
• NeurIPS Datasets & Benchmarks (2022): OpenSRH A public clinical dataset and benchmark for intraoperative brain tumor imaging.
• Work on a rare data problem at meaningful scale. You will organize deeply multimodal data generated across a major academic health system, not a small benchmark assembled for one paper.
• See your engineering work become science. The systems you build will enable new models, datasets, manuscripts, and clinical studies, with opportunities for authorship and technical leadership.
• Work alongside the clinical environment. Collaborate with physicians and scientists who understand how data are generated, where current AI fails, and what would improve patient care.
• Help define a new field. Health system learning requires new approaches to data architecture, multimodal learning, evaluation, governance, and agent design. This role will help shape those foundations.

Background Screening

Michigan Medicine conducts background screening and pre-employment drug testing on job candidates upon acceptance of a contingent job offer and may use a third party administrator to conduct background screenings. Background screenings are performed in compliance with the Fair Credit Report Act. Pre-employment drug testing applies to all selected candidates, including new or additional faculty and staff appointments, as well as transfers from other U-M campuses.

• Master's degree or higher in computer science, data science, computer engineering, biomedical engineering, bioinformatics, informatics, or a related field. Candidates with substantial equivalent professional experience are also encouraged to apply if permitted by the University job classification.
• Strong Python and SQL skills, with experience building production-quality data pipelines, databases, or scientific software.
• Fluency with Linux, Bash, Git, testing, debugging, documentation, and collaborative software-development practices.
• Experience working with large, heterogeneous datasets and designing reliable, maintainable, and reproducible systems.
• Experience with high-performance, distributed, or cloud computing; familiarity with SLURM is strongly valued.
• Ability to work independently and collaborate across disciplines, with a strong commitment to scientific rigor, data stewardship, responsible AI, and clear communication., * Experience with DICOM, PACS, whole-slide imaging clinical data warehouses, medical imaging, computational pathology, clinical NLP, or longitudinal EHR data.
• Experience with PyTorch and self-supervised learning, vision-language modeling, large language models, or foundation models.
• Experience with scalable data technologies such as Spark, Dask, Ray, dbt, Airflow, Prefect, or comparable systems.
• Thoughtful use of coding assistants and agents, such as Claude Code or Codex, combined with careful review, testing, security, and reproducibility.
• Research contributions, open-source software, technical leadership, or publications at leading machine learning conferences or biomedical journals.

Michigan Medicine improves the health of patients, populations and communities through excellence in education, patient care, community service, research and technology development, and through leadership activities in Michigan, nationally and internationally. Our mission is guided by our Strategic Principles and has three critical components; patient care, education and research that together enhance our contribution to society., Michigan Medicine is one of the largest health care complexes in the world and has been the site of many groundbreaking medical and technological advancements since the opening of the U-M Medical School in 1850. Michigan Medicine is comprised of over 30,000 employees and our vision is to attract, inspire, and develop outstanding people in medicine, sciences, and healthcare to become one of the world's most distinguished academic health systems. In some way, great or small, every person here helps to advance this world-class institution. Work at Michigan Medicine and become a victor for the greater good.