MindInventory partnered with Biostrap, now operating as Sensor Bio, working directly with CEO and Cofounder Sameer Sontakey on the engineering behind a wearable platform built for a use case consumer wearables cannot serve: giving researchers and clinicians access to the raw physiological signal, not just the number a device decided to display.
Most wearables compute a heart rate on the device and discard the waveform that produced it. Biostrap keeps it. That single architectural decision, preserving raw PPG, accelerometer and gyroscope data end to end and then processing it in the cloud rather than on a battery constrained wrist device, is what makes the platform usable in IRB approved studies and remote patient monitoring rather than fitness tracking. The platform is used by Stanford, UCLA, NASA, Pfizer, Tulane and the NIH, and today reports more than 400 million hours of captured data across 100 clinical studies.
Engineering support across Biostrap’s mobile app, cloud biosignal pipeline, and the HIPAA compliant dashboard used by clinical and research customers
Raw PPG, accelerometer and gyroscope waveform data preserved end to end, rather than compressed into derived metrics on the device
A configurable data, survey and app layer that lets one platform serve first responders, sickle cell researchers and elite athletes without a separate build for each
Used by Stanford, UCLA, NASA, Pfizer, Tulane, CHLA and the NIH, with more than 400 million hours of captured data across 100 clinical studies
now operating as
Sensor Bio
HIPAA compliant platform
Used in clinical research and remote patient monitoring
contexts
STANFORD
NASA
PFIZER
NIH
UCLA
400
Million hours of captured data
100
Clinical studies
HIPAA
COMPLIANT
Research and monitoring platform
Sameer Sontakey started Biostrap after losing close to 80 pounds and becoming seriously interested in his own physiology, then discovering that the wearables available to him were built to withhold exactly the data he wanted.
That is not a design oversight. It is the consumer wearable business model. A fitness tracker’s job is to show a clean number that makes the user feel informed, which means aggressive smoothing, gap filling, and proprietary processing that never leaves the device. It is the right call for a consumer product and it makes the data useless for research, because a researcher cannot distinguish a real physiological change from an artifact of someone else’s undisclosed filtering.
The people who needed that data had no good option. A university lab studying autonomic recovery, a clinician monitoring sickle cell patients between appointments, a health company trying to build a product on validated biometrics: all of them were choosing between a consumer device that hides its signal and building sensor infrastructure from scratch. Biostrap was built for the gap between those two. Research grade capture, raw data access, and a platform configurable enough that each of those very different users could run their own study on it.
Beat to beat precision, not averages
The metrics that matter clinically depend on millisecond accuracy in beat detection. Getting that from an optical sensor on a moving wrist is a materially harder problem than reporting an average heart rate.
HIPAA from the architecture up
Supporting remote patient monitoring and clinical research meant compliance had to shape the data model, not get audited onto it later.
One platform, incompatible use cases
Every enterprise and research customer wanted something specific. A custom build per customer does not survive the fifth customer.
Raw data access at clinical fidelity
Most wearables process and discard their raw sensor data. Researchers needed the underlying PPG and motion waveforms, timestamped and exportable, not a derived summary.
Configurability without forking
Every enterprise and research customer wanted something specific. A custom build per customer does not survive the fifth customer.
1
A validated biosensor captures raw photoplethysmography (PPG), accelerometer and gyroscope data continuously, at a sampling rate high enough to resolve individual pulse morphology, not just detect that a beat occurred.
2
Data moves off the device through the mobile app, which can be configured per customer for use case, branding and required survey instruments.
3
The cloud pipeline filters and cleans the waveform data, identifies and excludes motion corrupted segments, and derives biometrics from what remains. Raw data is preserved alongside the derived values rather than replaced by them.
4
Heart rate variability, respiratory rate, and autonomic nervous system balance are computed from the cleaned signal, measures that depend on the beat timing precision the capture layer was built to preserve.
5
Results surface through a HIPAA compliant dashboard for clinicians and researchers, or through APIs and SDKs for partners building their own products on validated data.
6
For research and enterprise customers, the platform supports the full validation path, from study protocol design through publication, so product claims rest on data rather than assertion.
Raw waveform access as a first class decision, not an
export feature
Preserving raw data end to end constrains everything upstream and downstream: sampling rate, on device storage, transmission bandwidth, cloud storage cost, and the data model itself. It is meaningfully more expensive than shipping derived metrics. It is also the only reason the platform can serve a researcher who needs to run their own analysis rather than trust someone else’s.
Motion artifact handled by exclusion, not
interpolation
A moving wrist produces optical noise that superficially resembles a pulse. The accelerometer and gyroscope streams are used as context for the PPG signal, so corrupted segments can be identified and dropped. Consumer devices tend to interpolate across them, because a gap in a chart looks like a broken product. In a study, an interpolated value is worse than a missing one. It is a fabricated data point that no downstream analysis can identify as such.
Processing in the cloud, not on the wrist
Waveform level analysis takes more compute than a wearable can spend without either draining the battery in hours or cutting corners on the signal. Splitting the concerns, so the device captures faithfully and the cloud processes, means the analysis can be improved or re run across historical data without shipping firmware. For a two year longitudinal study, consistent reprocessing of the full dataset is worth more than a marginally faster result on the device.
Configuration built into the platform, not forked per
customer
Data parameters, survey instruments and app experience are configurable on top of a shared capture and processing foundation. That is what lets first responders, sickle cell researchers and athletes run on one system, and what keeps the engineering cost of a new enterprise customer close to zero.
HIPAA in the data model
Compliance shapes access control, audit logging, retention, and how identifiers are separated from physiological data. Those decisions sit underneath the first chart ever drawn. Retrofitting them means rebuilding the data model, which is why they were made first.
A platform used by NASA, Pfizer and university research teams clears a different bar than a consumer fitness app. Biostrap’s device and algorithms have been independently validated against gold standard references, including simultaneous ECG recording under laboratory protocols. Validation is scoped to the conditions actually tested, because performing well at rest says nothing about performance during exercise.
Published research using the platform spans detecting COVID 19 symptom onset in biometric trends to monitoring pain episodes in sickle cell patients. That kind of work only holds up if data provenance, processing steps and methodology stay transparent enough for peer review, which is a constraint on the architecture as much as on the science.
400
million hours of captured data
Longitudinal capture at genuine scale, not pilot sized datasets
100
clinical studies
Real research adoption, with published outcomes rather than stated intent
Stanford, UCLA, NASA,
Pfizer, Tulane, CHLA, NIH
Named, checkable institutional customers
Validated
against ECG reference
Independently benchmarked against the clinical gold standard
Consumer Wearables
Biostrap
Derived metrics only
Raw PPG, accelerometer and gyroscope waveforms, timestamped
and exportable
Smoothed or interpolated to keep charts continuous
Identified and excluded, so downstream analysis knows what is
missing
Fixed experience for every user
Configurable sensors, parameters, surveys and app per use
case
Not built for clinical use
HIPAA compliant, used in published clinical research
MindInventory worked alongside Biostrap’s CEO on the platform’s mobile application, the cloud data pipeline that processes raw biosensor waveform data at scale, and the HIPAA compliant web dashboard used by clinical and research customers.
Photoplethysmography measures blood volume changes by shining light into the skin and reading what comes back. Every optical wearable uses it. The difference is what happens next. A consumer device runs the signal through proprietary filtering on the device, outputs a heart rate, and throws the waveform away. A researcher studying autonomic function needs the waveform itself, the morphology of each pulse, the exact timing between beats, the noise characteristics, because that is where the signal they are investigating actually lives. Once it has been averaged into a number, it cannot be recovered.
Heart rate tolerates error. If a beat is detected 40 milliseconds late, the average barely moves. Heart rate variability is built entirely from those inter beat intervals, so a 40 millisecond timing error is not noise around the measurement. It is the measurement. That means beat detection has to be precise at the millisecond level, which requires a high sampling rate, a clean signal, and detection logic that does not quietly interpolate over beats it is not sure about. Most consumer variability numbers come from intermittent sampling and heavy smoothing, which is why they do not hold up in research.
Motion is the dominant failure mode in optical sensing. A wrist moving during a walk produces optical noise that looks superficially like a pulse. Solving it means using the accelerometer and gyroscope streams as context for the PPG signal rather than as separate metrics, so the pipeline can identify which segments are corrupted and exclude them rather than average through them. The honest approach is discarding bad data and saying so. Consumer devices tend to interpolate instead, because a gap in the chart looks like a broken product. In a clinical study, an interpolated value is worse than a missing one.
Battery and compute. Meaningful waveform level analysis takes more processing than a wrist worn device can do without either draining in hours or cutting corners on the signal. Streaming raw data off the device and processing it in a dedicated cloud pipeline separates the two concerns. The wearable’s job is faithful capture, the cloud’s job is filtering, cleaning and deriving biometrics. It also means the analysis can be improved, or re run against historical data, without shipping new firmware, which matters when a study needs consistent processing across two years of collection.
By building configuration into the platform rather than forking it. Each of those groups needs different data parameters, different survey instruments, and often a differently branded mobile experience, but they all sit on the same capture and processing foundation. The alternative, a custom build per customer, does not survive the fifth customer. Making the data layer, the survey layer and the app experience configurable is what turns a product into a platform.
More than encryption. It shapes access control, audit logging, data retention, and how identifiers are separated from physiological data. Those decisions have to sit in the architecture before the first chart gets drawn, because retrofitting them means rebuilding the data model. For a platform supporting remote patient monitoring and clinical research, compliance is a precondition for the customer conversation happening at all, not a feature that differentiates it.
Against a gold standard reference under controlled conditions. For cardiac measures, that typically means simultaneous ECG recording across a range of activity levels and physiological states, then statistical comparison of the two. Biostrap’s measurements have been independently validated against references including ECG chest straps and laboratory protocols. Validation is also specific to the use case. Performing well at rest says nothing about performance during exercise, which is why claims have to be scoped to the conditions actually tested.
Research and clinical organizations including Stanford, UCLA, NASA, Pfizer, Tulane, CHLA and the NIH, alongside enterprise customers such as Therabody, Pepsi, Gatorade and Clorox. Published research using the platform spans detecting COVID 19 symptom onset in biometric trends to monitoring pain episodes in sickle cell patients.
Yes. We work as a product engineering partner on connected health products: biosensor data pipelines, mobile applications for continuous capture, HIPAA compliant clinician dashboards, and the API and SDK layers that let partners build on top of validated data. Biostrap represents the demanding end of that, with research grade fidelity, regulated data handling, and one platform serving clinical, enterprise and consumer contexts at once. Our other healthcare work includes Sully AI’s clinical AI workforce and Arrow’s agentic AI platform for revenue cycle management.
