Guhan Sundar

EEG · MEG · WEARABLES · DATA INFRA — 2012 → NOW

Instrumenting the body

research · 2019-10 → ONGOING · 6y 10m · CROSS-ERA

The conviction that good data is the bottleneck — from MEG sensors to clinical EEG to everyday health capture.

KernelAlto NeuroscienceIndependent

What

One belief, held for a decade and pursued at three scales: understanding the body is a data problem, and almost everyone underinvests in the data itself.

Kernel aimed it at the brain at the source — a whole-head, 432-magnetometer optically-pumped MEG system built to record while a person moves and behaves naturally. Alto aimed it at clinical biomarkers — EEG, ERP, wearable, speech, and cognitive signals collected at psychiatric trial sites and turned into numbers a drug-development team can act on. The independent work aims it at the whole real-world human, where nobody is wearing a research-grade cap and the data has to be earned from an ordinary day.

Why it mattered

The belief crystallised at Imperial. Understanding the brain requires good data, and there was no good non-invasive, real-world way to get it: EEG is noisy and confined to artificial laboratory tasks. That is what settled the PhD-versus-industry question — the real world was the interesting problem, and Kernel was attacking it at the source.

Each attempt then taught the next one its constraint. Clean MEG needs a magnetically shielded room, which is a fundamental scaling problem for anything field-deployable. Clinical EEG buys its quality with a protocol, which means only ever measuring people inside a clinic. Real-world capture has no protocol to hide behind, which makes the design problem motivational as much as technical: if the capture isn’t enjoyable, there is no data.

My role

Every station of the pipeline, in each era: hardware design, assembly, and calibration; acquisition environments and site onboarding; preprocessing, artifact rejection, and QC; statistics and machine learning; and now LLM tooling layered on top. The range isn’t scatter — building instrumentation is what makes it necessary.

Outcome

At Kernel, a 432-magnetometer OP-MEG system taken from early prototype to lab-ready instrument, and a co-authored SPIE paper describing it. At Alto, an ERP biomarker program built from nothing, and EEG biomarkers used in Phase 2 trials for both major depressive disorder and schizophrenia. Independently, capture tools shipped and in daily use.