How a Smartphone Replaced the Urologist's Office — A Medical-Grade App That Turns Audio Into Clinical Data
The Problem
EmanoFlow had a patented ML model that converts urination audio into clinical-grade flow measurements — no hardware, just a smartphone. The science was validated. But turning it into a HIPAA-compliant, cross-platform mobile app that elderly patients could reliably use at home required healthcare-grade engineering, not general app development. The bar for clinical precision and regulatory compliance demanded deep domain expertise in both mobile and healthcare.
Why Generic Mobile Teams Fail Here
A smartphone microphone is not a medical instrument — and clinical decisions depend on the output. Turning a validated ML model into a tool elderly patients use reliably at home demands healthcare-grade engineering where consumer-grade mobile development breaks down:
- A smartphone microphone is not a medical instrument — device variation across hundreds of models, bathroom acoustics, and background noise all degrade the signal, and clinical decisions depend on the output
- HIPAA compliance end-to-end — every recording, transmission, and storage step must be encrypted and auditable across the full pipeline
- Clinical equivalence across iOS and Android — the app must produce clinically equivalent results despite different microphone hardware, audio APIs, and permission models
- High compliance from elderly patients with limited tech experience — the UX must drive 30-75 completed recordings per order from a demographic where activation barriers kill adoption
What We Did
Clinical-Grade Mobile Architecture
- Built the cross-platform application on React Native with Expo/EAS, delivering native iOS and Android from a single codebase — critical for a startup reaching both platforms without doubling engineering cost
- Designed the audio capture module to maximize recording fidelity across device variations — standardized sampling parameters, device-specific calibration, and pre-recording environment checks (ambient noise detection, mic access verification)
- Architected the HIPAA-compliant data pipeline (on-device encryption, secure transmission to the ML API, encrypted AWS storage on Kubernetes) with an automated testing framework across a matrix of device models and OS versions
Patient & Provider Applications
- Engineered the patient recording experience for minimal cognitive load — large tap targets, a step-by-step guided flow, real-time audio quality feedback, plain-language result summaries, and automatic day/night separation for nocturia analysis
- Integrated IPSS (International Prostate Symptom Score) surveys and trend visualization across recordings, driving the 30-75 recordings-per-order completion rates
- Built the provider app and web portal for creating measurement orders, reviewing longitudinal data, and generating CPT-billable clinical reports, with SMS-based patient onboarding — a text, a tap, and recording starts, with zero app-store friction
ML Integration & Clinical Reliability
- Integrated the proprietary ML API into the mobile pipeline — audio capture flows seamlessly into ML processing and returns structured clinical data (flow rate in ml/s, voided volume in ml, and complete flow-curve visualizations)
- Built real-time feedback loops that tell the patient whether the measurement succeeded, reducing wasted recordings and improving clinical data yield across the population
- Implemented comprehensive edge-case error handling and engineered the AWS infrastructure (Kubernetes, auto-scaling) for unpredictable at-home usage spikes while maintaining HIPAA compliance across the stack
Key Results
In Their Words
What really differentiates them is their ability to deeply understand business needs.
Their proactive team gets things done as if it were their own project.
What We Learned
Audio capture for medical data is fundamentally different from consumer audio
A smartphone microphone was never designed as a clinical instrument. Device-to-device variation in sensitivity, sampling behavior, and audio processing means a recording good enough for a voice memo may be useless for clinical measurement. A medical-grade capture module requires standardized sampling, device-specific calibration, and pre-recording environment checks — none of which exist in off-the-shelf audio libraries. If your algorithm depends on audio input, the capture engineering is as critical as the ML model.
90%+ elderly patient compliance is an engineering outcome, not a design outcome
When users are 85 years old and clinical value depends on completing 30-75 recordings per order, UX is not about aesthetics. It is about engineering every interaction so the interface becomes invisible: guided flows with real-time audio feedback, automatic day/night separation, plain-language results, and SMS-based onboarding. When UX determines whether a surgeon has usable data, it stops being a design problem and becomes an engineering problem.
Cross-platform clinical equivalence requires device-specific calibration
iOS and Android handle audio capture differently — APIs, microphone hardware, background processing, and permissions all vary. A React Native codebase that works identically for consumer audio does not produce clinically equivalent results for medical measurement. Achieving equivalence required device-specific calibration, automated testing across a device matrix, and a continuous validation pipeline so platform updates never break clinical accuracy.
Have a Clinical Algorithm to Ship?
Have a validated clinical algorithm that needs to become a patient-facing mobile app? We took EmanoFlow from validated ML model to deployment — including the audio capture engineering, HIPAA pipeline, and elderly-user UX behind 90%+ compliance. Book a 30-minute session. No pitch deck. Just engineering clarity.