Pre-clearance research prototype

Contactless patient vitals from $50 of WiFi hardware.

ViFi is building an FDA 510(k)-track contactless patient monitor that runs on commodity ESP32 chips. We've shown 4.15 bpm cross-session heart-rate accuracy on real hardware against a chest-strap reference — at roughly 1/20th the cost of the bedside monitors hospitals deploy today.

Pilot inquiry → Read the methodology
HR MAE (cross-session)
4.15 bpm
Hardware cost / room
~$50
Validation
LOSO × 4

leave-one-session-out

Ground truth
Polar H10

chest-strap, BLE

The problem

Hospital patients outside the ICU are unmonitored more than 23 hours per day.

On general medical/surgical floors — roughly 700,000 U.S. beds — the standard of care is a vitals reading every 4 to 8 hours by a nurse. Between those readings, transient deteriorations go unobserved: the fever spikes of brewing sepsis, the heart-rate surges of a bacteremia flare, the respiratory depression that precedes an opioid-related code.

Existing continuous-monitoring solutions cost roughly $3,000 per bed and require wires, electrodes, or wearables that patients remove, dislodge, or refuse. They're reserved for the ICU and step-down units. The 80% of beds where most preventable deteriorations actually happen get nothing.

ViFi exists because the founder's mother spent 31 days getting diagnosed with Staphylococcus bacteremia after four separate visits, each time with vitals "stable" by the moment the nurse arrived.

How it works

WiFi packets bend around bodies. We listen to how.

Two ESP32-S3 chips — one transmitting, one receiving — exchange WiFi packets at roughly 70–100 packets per second. Each received packet's per-subcarrier amplitude (Channel State Information, or CSI) reflects the multipath environment in the room, including chest-wall motion from breathing and heartbeat. No line of sight, no contact, no patient compliance. 

[ESP32-S3 TX] ─────── 192 subcarriers ───────► [ESP32-S3 RX]
   antenna          (chest perturbs path)         antenna
                                                     │
                                                     │  USB serial @ 921600 baud
                                                     ▼
                                            csi_capture.py
                                                     │
                                                     ▼
                                       first_capture_report.py
                                                     │
                                                     ▼
                  HR / RR predictions  vs  Polar H10 ground truth

Signal path

  1. Variance-rank top-K subcarriers
  2. Butterworth 0.1–3 Hz bandpass
  3. 4× zero-padded FFT
  4. Parabolic peak refinement (cardiac and respiratory bands)
  5. 9-dimensional feature vector → XGBoost regressor

Where the science comes from

The signal-processing approach is from peer-reviewed academic work — PhaseBeat (INFOCOM 2017), FullBreathe (UbiComp 2018), ResBeat (2020). ViFi's contribution is productizing it on commodity ESP32-S3 hardware, with calibration, RF fingerprinting, out-of-distribution suppression, and an auditable prediction pipeline suitable for an FDA submission.

What we've shown

4.15 bpm cross-session HR accuracy on commodity hardware.

Four paired captures, single subject, single room. Each capture pairs raw CSI from a 2-node ESP32-S3 array with continuous heart rate from a Polar H10 chest strap as ground truth.

For each holdout we trained on the remaining sessions and scored the held-out one — a leave-one-session-out evaluation that estimates how the system generalizes to a session it has never seen.

Mean absolute error across the two scored holdouts: 3.89 and 4.41 bpm (mean 4.15). Per-window accuracy within ±5 bpm: 65–68%. PhaseBeat reported 1.5 bpm using $500 Intel 5300 NICs in the same family of methods.

Full methodology, captures, and per-session breakdowns: RESULTS.md

[ photo of ESP32-S3 array + result chart goes here ]
System HR MAE Hardware Per-node cost
ViFi 4.15 bpm 2× ESP32-S3 ~$25
PhaseBeat 1.5 bpm 2× Intel 5300 NIC ~$500
Bedside monitor ≤ 1 bpm ECG / SpO₂ ~$3,000

What we're building

FDA 510(k) Class II contactless vitals monitor.

The current system is a pre-clearance research prototype, not a medical device. Our target is FDA 510(k) Class II clearance for heart rate and respiratory rate at roughly 1/20th the per-bed cost of bedside monitors. The path:

  1. 1

    Multi-subject + multi-room validation

    10+ subjects, 3+ rooms, varied HR ranges, paired RR captures with a Vernier respiration belt. Target: cross-subject HR MAE under 3 bpm.

  2. 2

    Wellness-grade hospital pilot (pre-FDA)

    5–10 medical/surgical beds, parallel to existing standard-of-care monitoring, fully de-identified. No clinical decision-making is made from ViFi data during the pilot.

  3. 3

    FDA 510(k) Class II submission

    ISO 13485 quality system, clinical validation study at an academic medical center, predicate-device comparison filing.

Live shipped-vs-planned manifest: ROADMAP.md

For hospitals

We're recruiting pilot sites.

We're talking to medical/surgical floor leaders, biomedical engineering teams, and clinical research offices interested in hosting a wellness-grade pilot of contactless monitoring.

A pilot looks like this:

  • 5–10 rooms instrumented with 2 ESP32-S3 boards each
  • 30–90 day duration, IRB review where applicable
  • Parallel to your existing standard-of-care monitoring; no clinical decisions are made from ViFi data
  • Fully de-identified, on-prem data handling
  • Output: a measurement study comparing ViFi against your existing reference
Email pilots@vifi.health →

For everyone else

The code is open.

ViFi is pre-incorporation. The DSP pipeline, training scripts, API, dashboard, calibration, RF fingerprint, audit log, and 102-test suite all live on GitHub. If you're a researcher, an engineer curious about CSI-based sensing, a press contact, or just want to know whether a $50 sensor really hits 4.15 bpm cross-session — read the code and the methodology.

GitHub → RESULTS.md ROADMAP.md zach@vifi.health