Academic oncology labs extract hundreds of thousands of clinical data points by hand every year. A single 200-patient retrospective study requires reviewing roughly 30,000 cells of data — labs, imaging, notes, treatment histories — populated into Excel and REDCap by residents, fellows, and undergraduates over months of repetitive work. Cohortly does this work as a service: an AI agent navigates the chart, interprets per protocol rules, and produces every value with a citation; the senior reviewer audits and signs off.
The Problem
Manual chart abstraction is the slowest layer of academic clinical research. The work is repetitive, requires interpretive judgment that residents and undergraduates aren't trained for, and consumes time that should be spent on hypothesis generation and analysis. It is also the layer most resistant to automation — autonomous AI extraction tools have been built at multiple academic medical centers and consistently fail to gain adoption, because clinical research data has an asymmetric failure mode where one wrong field can sink a publication.
We build for the human, not around them. Every Cohortly suggestion comes with a complete evidence trail — the source documents, the exact snippets consulted, the protocol rule applied. The reviewer verifies and signs off. Every override becomes a permanent institutional rule the agent applies on every future paper at that lab.
What It Is
Cohortly is an AI agent that operates the EHR for clinical research teams. When a researcher opens a CRF or registry field, the agent navigates the patient's chart in real time — switches tabs, opens notes, scrolls to imaging reports, highlights specific snippets — and produces a suggested value with full provenance. The researcher accepts, overrides, or asks for clarification.
We deploy locally on institutional hardware so protected health information never leaves the firewall. Audit trails meet 21 CFR Part 11 standards. We're piloting at Huntsman Cancer Institute under the genitourinary oncology program.
Who We Are
Cohortly is built by Krishnam Goel, Mingchuan Cheng, and David Zhang. Krishnam works as a research student in the GU oncology program at Huntsman Cancer Institute, where he has personally extracted approximately ten thousand chart fields by hand. Mingchuan studies at Stanford and researches reinforcement learning and agents at the Stanford AI Lab. David studies computer science at Columbia.
Contact
For research collaborations, pilots, or investment inquiries: krishnamgoel.9675@gmail.com