Applications open · Summer 2026

Apply machine learning
to open clinical questions.

Tensor Lab is a ten week summer research fellowship for machine learning engineers. You pair with a medical lead and a faculty physician at UCSF or UMSOM, own a scoped clinical research project from data to results, and leave with work you can present at the symposium, write up as a paper, or extend after the program.

View the 13 projects → Apply now

Scoped projects. Institutional data. End-to-end ownership.

Every project in the catalog exists because a practicing physician asked for it, a medical student scoped it, and a faculty PI agreed to sponsor the work. You come in as the engineer who builds it.

01

Clinician-sourced questions

Each of the thirteen 2026 projects was proposed by a practicing physician and scoped with a medical student partner. Work happens on institutional data: MIMIC, All of Us, UCSF Clarity, SEER, and other clinical sources.

02

Faculty mentorship

A faculty PI at UCSF or UMSOM sponsors every project. Weekly team check-ins, IRB and data access handled ahead of the sprint, and a direct line to clinicians who work on these problems full-time.

03

Own the research

You take a scoped clinical question from data to results over the ten week sprint. The summer closes with the Tensor Lab symposium, where each team presents their work to the cohort, faculty PIs, and invited clinicians.

You build. A medical lead guides. A physician sponsors the work.

Every project runs as a team of three. You write the code and run the experiments. A medical student lead translates the clinical question and keeps the work grounded. A faculty physician sets the scientific standard.

You are here

Machine Learning Fellow

You own the build. Training code, experiments, data wrangling, results write-up. Fifteen to twenty hours per week from June through August, remote. You are the technical lead on the project.

  • Strong Python and PyTorch or equivalent
  • Comfortable reading medical literature with support
  • Aiming for publishable work, not just a prototype
Apply to the 2026 Fellowship
Your medical lead

Medical Student Mentor

A medical student who framed the research question and helps you interpret results. Meets with you weekly and translates between the clinic and the model.

Your faculty PI

Physician Principal Investigator

A practicing physician and published researcher who sponsors the project, secures data access, and sets the scientific standard. Biweekly check-ins with the team.

What you take home from the summer.

Ten weeks of work, and four things that stay with you afterward.

  • A codebase you authored, open-sourced under your name on GitHub.
  • A symposium presentation delivered to the full cohort, faculty PIs, and invited clinicians.
  • Named authorship on any publication that comes out of the work, proportional to contribution.
  • A reference letter from your faculty PI on request.

From application to symposium.

Key dates for Summer 2026. Waitlist to final symposium in under five months.

Now

Waitlist signup

The waitlist is live on LinkedIn and through our informal channels. Add your name to get first notice when the full application opens on April 27.

April 27, 2026

Applications open

The application goes live alongside the campus listservs. Browse the thirteen projects, pick your top three, and submit one application. Plan for about thirty minutes.

May 26, 2026

Applications close

Submissions cut off. Matched fellows move into onboarding; data access setup begins the same day.

May 26 through June 16, 2026

Onboarding and data access

IRB paperwork, credentialing, platform access, and background reading. The logistics get handled up front so the sprint starts on research, not setup.

June 16 through August 25, 2026

Ten week research sprint

The fellowship kicks off June 16. Build the model, run the experiments, draft the write-up. Weekly technical reviews, biweekly faculty check-ins, one mid-sprint demo.

Tuesday, August 25, 2026

Final symposium

Each team presents their work to the cohort, faculty PIs, and invited clinicians. Manuscript drafting continues from there for teams ready to write up.

What last year's fellows built.

Seven projects from our 2025 pilot cohort. Every team shipped final results at the summer symposium.

Benchmarking · ICUNCI

LLMs vs. traditional ML

Compared LLM based semantic models against XGBoost across clinical prediction tasks.

Result. Text representation critically affects performance. Traditional ML still wins on pure physiological signals.
PharmacogenomicsUCSF

Oncology knowledge graph

Built an ETL first prototype for clinical evidence synthesis using a Knowledge Graph RAG system.

Result. Validated ETL first strategy and end to end KG RAG functionality on real oncology cases.
Oncology · SDOHUCSF

Prostate cancer and SDOH

Used SEER registry data linked with social determinants of health proxies to stratify prostate cancer patients.

Result. SDOH adds measurable predictive value, highlighting social context in risk models.
Head and Neck CancerUMSOM

RFS prediction via PET/CT

Investigated whether a simplified multimodal model using only PET/CT and clinical data could match expert annotated performance.

Result. Simpler model matched performance and generalized better across institutions.
Emergency MedicineKaiser

LLM ER triage simulator

Built a training simulator using LLMs to create interactive, data grounded emergency triage encounters from MIMIC IV.

Result. Confirmed feasibility of LLMs for realistic medical training tools.
Cardiology · ReadmissionsUCSF

Predicting readmissions

Attention based deep learning over free text discharge summaries to predict thirty day readmissions.

Result. AUROC 0.730 with interpretable predictions via SHAP.
Lung CancerUCSF

Deep learning for CT lung

Compared ResNet50, VGG16, and InceptionV3 for CT based lung lesion classification on the IQ OTH NCCD dataset.

Result. InceptionV3 showed the most consistent performance with 0.92 F1.

Fellow stories.

Tensor Lab showed me meaningful challenges that combine both clinical and technical domains. As I prepare for my PhD applications, this experience was invaluable in clearly defining my research direction.

Jialong Research Fellow, 2025

Tensor Lab taught me to conceptualize and develop applications for LLMs that require significant domain knowledge.

Ashwin Research Fellow, 2025

The guidance from the faculty and mentors was exceptional. I wouldn't have reached our final publication goals without their help.

Koushik Research Fellow, 2025

Collaborating with a global cohort provided me with a network of contacts for future projects.

Juan Arturo Research Fellow, 2025

The team behind the fellowship.

Two founders and three additional directors, all current medical students, running the fellowship end to end.

Founding Directors

Matt Allen Matt Allen Co-Founder · Executive Director UCSF School of Medicine
Aaron Ge Aaron Ge Co-Founder · Technical Director University of Maryland School of Medicine

Additional Directors

Chy Murali Chy Murali Operational Director University of Maryland School of Medicine
Gavin Shu Gavin Shu Strategic Director UCSF School of Medicine
Angie Lee Angie Lee Business Director University of Maryland School of Medicine

Common questions from fellows.

Do I need a medical background to apply?

No. Your medical lead and faculty PI provide clinical context. We are looking for strong Python, solid ML fundamentals, and curiosity about medicine. You will pick up the clinical side as you go.

How many hours per week during the sprint?

Plan for fifteen to twenty hours a week from June through August. Think of it like a serious research internship. If your circumstances shift, we can adjust.

Is this remote or in person?

Remote by default. You meet your team over Zoom or Slack. Some chapters organize optional in person working sessions if there is geographic overlap.

Is this paid?

No. The fellowship is unpaid.

Who owns the code and the research?

We default to open science. Code is open sourced and all team members share authorship. Patient data stays under the PI's institutional controls.

What if I am not a student?

We accept applications from current students, recent graduates, and career switchers with strong ML skills. The fellowship runs in the summer, but participation is not limited to people on academic calendars.

What happens if my top choice project gets filled?

You rank three projects when you apply. If your first choice is filled before your application is processed, you stay in the running for your second and third choices. You will receive an email with a link to swap in a new choice if one of yours is filled.

Summer 2026 Fellowship · Applications Open

Ten weeks of clinical ML research.

Thirteen projects across UCSF and UMSOM. Pick three, rank them, and apply in about thirty minutes. We follow up with next steps shortly after the deadline.