From Ubicomp to Clinical AI: My Research Pivot

People sometimes ask why a ubicomp researcher is now working on chronic pain. The short answer: the problems changed, but the values didn’t.

Ubiquitous Problems, Ubiquitous Solutions

My PhD work on privacy-preserving sensing taught me something beyond the specific technical contributions: the most meaningful systems are the ones that respect the people they’re designed for. PrivacyMic, PrivacyLens, SAWSense—each was built around the idea that sensing doesn’t have to mean surveillance. And these sensors had immediate clinical applications: PrivacyMic enabled privacy-preserving voiding monitoring and SAWSense demonstrated wearable acoustic ECG sensing—both published at EMBC—while PrivacyLens is now being used with clinical colleagues at UM for robust fall detection.

Pain research has that same tension at its core. Over 50 million Americans live with chronic pain, yet assessment still relies almost entirely on self-report. We need better measurement, but “better” can’t come at the cost of patient dignity or autonomy. And if we’re going to address a problem this ubiquitous, we need solutions that work continuously, unobtrusively, and without adding to the burden of being sick—computers that, as Mark Weiser envisioned, fade into the background, present only when needed.

What I’m Working On

At Stanford’s SNAPL, our work is organized around three questions: What causes acute pain to become chronic after injury or surgery? Can we predict who’s at risk early enough to prevent chronification? And for those already living with chronic pain, what treatments actually work?

What drives me is that first question—prediction. If we can identify risk early enough, we have a chance to intervene before someone’s pain becomes a permanent part of their life. That reality never has to happen. And the computers woven into the fabric of our daily lives—smartwatches, phones, wearables—are already there, ready and waiting to do their part. They’re the cornerstone of prediction.

But to forecast accurately, we need to understand the underlying signals. Can we trace it back to interpretable biomarkers? Is it something in the way people move—captured through actigraphy—or patterns in brain activity visible on fMRI? I’m exploring multimodal approaches to identify these biomarkers, drawing on a basic human intuition: we can often tell when someone is in pain just by watching them walk. If humans can do this intuitively, can we train smartwatches—something already on millions of wrists—to do the same?

This requires integrating clinical records, patient-reported outcomes, wearable data, and social determinants of health. The models need to be accurate, but also fair, interpretable, and privacy-preserving.

Specific directions include:

• Personalized treatment response prediction—moving beyond one-size-fits-all pain protocols
• Multimodal biomarker discovery—using actigraphy, fMRI, and other sensing modalities to find interpretable signals
• Continuous, privacy-preserving pain sensing—leveraging smartwatches and everyday devices, applying the same design philosophy from my PhD
• Bias mitigation—ensuring algorithms work equally well across patient populations without amplifying existing disparities in pain care
• Interpretable decision support—models clinicians can actually trust and use

What Connects It All

The tools are different—magnetometers to EHR data, desk surfaces to clinical workflows—but the design philosophy is the same: build systems that serve people, not the other way around. Embed constraints that protect users. Design for the real world, not the lab.

As an undergrad, I started in biomedical engineering because I wanted to go to med school. Being a first generation American, I’d been taught that medicine was the path you took if you wanted to help people. I never made it to med school—unless an Epidemiology degree from Stanford Med counts, which I’m told it does not—but in a way, I’ve come full circle. The problems I work on now are the same ones that drew me to medicine in the first place: fifty million people in the U.S. live with chronic pain, and most of us know someone who does (and if you’re reading this and know me, now you definitely know someone who does). This work is personal, and that’s exactly why it matters.