Home intelligence.
Three layers deep.


Know before it breaks.

HVAC fault detection · Edge AI · Plain-English diagnosis

See Our Products

Protect your biggest home system.


Sensee builds hardware-plus-AI products that give homeowners real visibility into the systems that matter most — starting with your heat pump and AC.

HVAC Vibration Guard

ESP32 + Accelerometer + Gemini AI

Mounts magnetically to your heat pump or AC unit. Three independent intelligence layers — edge signal processing, Android adaptive AI, and Gemini language diagnosis — detect mechanical faults weeks before they become failures.

  • 400 Hz FFT on 5 diagnostic frequency bins
  • 8 specific fault types identified by name
  • Natural language diagnosis via Gemini AI
  • No subscription. Ships assembled and ready to mount.
Learn More

How HVAC Guard works


Three independent intelligence layers, each catching what the others might miss.

Layer 1

ESP32 Edge Processing

The LSM303D accelerometer samples at 400 Hz directly on the unit. An onboard FFT resolves vibration into five diagnostic frequency bins (33, 60, 66, 99, and 120 Hz), computing per-axis RMS and peak values in real time. A state machine tracks the compressor lifecycle — OFF, STARTING, RUNNING, STOPPING — so every measurement carries operating context.

400 Hz sampling 5-bin FFT Per-axis RMS State machine
Layer 2

Android Adaptive AI

The companion Android app receives the edge data and applies statistical models calibrated to your specific unit. An EMA adaptive baseline adjusts to seasonal drift, while CUSUM drift detection and Mahalanobis distance analysis on the full FFT vector catch deviations that would otherwise be hidden in noise. Eight targeted fault classifiers compare the current spectral signature against known mechanical failure patterns.

EMA baseline CUSUM drift Mahalanobis distance 8 fault classifiers Rate-of-change regression
Layer 3

Gemini AI Diagnosis

When Layer 2 detects a fault state change, Gemini receives the full diagnostic context — frequency bin readings, fault type, severity trend, and operating history — and produces a plain-English explanation of what is likely happening and what to do about it. A persistent chat interface lets the homeowner ask follow-up questions in natural language without any technical knowledge required.

Natural language diagnosis Fault-triggered alerts Homeowner chat interface

Eight fault types. Named, not guessed.


Every alert comes with a specific mechanical cause — not a generic warning. The system knows the difference between a failing bearing and a loose panel, and tells you so.

Fan Imbalance

Blade-pass frequency elevation indicating a bent blade, debris accumulation, or imbalanced fan assembly.

Bearing Wear

High-frequency harmonic rise combined with CUSUM drift — the earliest detectable sign of bearing degradation.

Debris / Obstruction

Intermittent crest factor spikes revealing contact between a rotating component and foreign material.

Panel Rattle

99 Hz and 120 Hz resonance peaks identifying loose sheet metal or cover panels vibrating at grid harmonics.

Coil Icing

Compressor frequency bin shift combined with sustained trend — detects ice formation that chokes airflow and strains the motor.

Startup Stress

Abnormal startup transient profile — identifies hard-start conditions that shorten compressor life with each cycle.

Spectral Anomaly

Mahalanobis multi-bin shape deviation — catches unusual spectral fingerprints that don't match any named fault but signal something has changed.

Elevated Vibration

Sustained RMS trend exceeding adaptive baseline — a general mechanical stress indicator when no specific fault matches.

Get in touch


Interested in early access, partnership, or just have a question? We read every message.