All case studies
Research

Cancer research analysis driven by the biology, not the defaults

Working with a Stanford hematology fellow to analyze flow cytometry data with methods chosen by the science, then presenting the work at NYSDS 2026.

Dr. Gabriel Salzman, a hematology and oncology fellow at Stanford, works with spectral flow cytometry, which measures dozens of markers across millions of cells. Datasets that rich are still usually analyzed with the same manual gating used for classical flow cytometry. Computational methods that explore far more of the data are published, but implementing even one well is real work, and each method sees the data its own way.

Dr. Salzman chose the questions and decided how we would know an answer was right: any population the analysis flagged had to be reproducible by hand in FlowJo from the same raw data. With that rule in place, we directed AI agents to search the literature for suitable clustering and population-discovery approaches, implement and tune each one, and build a dashboard for comparing what they found. Agents also helped train a model that draws gates automatically.

Illustration with invented data. A standard dot plot of Marker A against Marker B shows every cell blurred together. A prompt adds a cell map colored by cluster, linked to the dot plot, with cluster C3 selected. A second prompt splits C3 into subclusters and plots subcluster SC2 on its own, where two populations that were blurred together separate. In the agent-built versions, hovering a cell highlights it in every linked view.

Standard tool

Prompt

Add a map of these cells colored by cluster, linked to the dot plot.

Prompt

Split cluster C3 into subclusters and plot Marker A vs B for SC2.

Along the way, a custom view built during the analysis showed that a population of interest was being blurred together with phenotypically distinct cells, something the summary statistics hadn't revealed. That pushed us to broaden the set of methods, and together they surfaced clearer hypothesis candidates than any single one. Early on I steered the agents. Over time Dr. Salzman began to take the lead instructing the agent, and today he extends the analysis himself.

The work became a poster, “Dynamic AI Agent-Built Visualization Tools for Scientific Inquiry,” at the New York Scientific Data Summit (NYSDS) 2026, which I presented as first author.

More case studies

  • Analysis at scale for a nonprofit research team, in a secure environment

    Six months with the analytics team at the Center for the Helping Professions, a nonprofit working in partnership with the University of Kentucky, building a language-model classification workflow their team now extends on its own.

  • Replacing the spreadsheet that ran the planning process

    Working with a program management leader at a large tech company to replace the spreadsheet planning process their executives reviewed monthly, with a tool they built, deployed, and extended themselves.

  • From retouching by hand to building a design org's tools

    Coaching a professional image-retouching expert, from a first prototype to building and maintaining AI image tools used across their design organization.

View all case studies

I build AI fluency by working beside people on their actual work.

Stop experimenting with tools. Start solving problems. Every one of these stories started with a real one. Tell me a little about yours, or your team's, and we'll find a useful first project on a short call.

Book a discovery call
dan@wvlen.com