Inside MoldSentrix

MoldSentrix is often described as "smart humidity sensors," but that undersells what's running behind the scenes. This article walks through the actual architecture — what the sensors measure, where the data goes, and how the AI decides that a particular closet is two weeks away from a mold problem.

Step 1: Sensors That Never Sleep

During installation, we place compact wireless temperature and humidity sensors in a home's highest-risk zones — bathrooms, kitchens, bedrooms, closets on exterior walls. The sensors transmit their readings to a small base station in the home, and the station forwards them to our monitoring server every few minutes.

Each reading arrives tagged with the exact sensor and room it came from. If a sensor goes quiet, we notice — staleness is tracked, so a dead battery doesn't silently become a blind spot.

Step 2: A Time-Series Memory of Your Home

Every reading lands in a time-series database, tied to a specific sensor, room, and property. We keep a rolling 90-day history — enough to see not just "what is the humidity now" but "how has this room behaved across weeks of weather, AC cycles, and daily routines."

Alongside the readings, the system stores installation context: the property's address and climate, construction notes, the HVAC schedule, and per-room details like "sealed storage closet, no ventilation" versus "bathroom with exhaust fan." This context turns out to be the difference between a useful alert and noise.

Step 3: AI Analysis That Pulls Its Own Data

Once a day, an AI analyst — built on Anthropic's Claude — reviews every monitored property. This isn't a fixed formula. The model works through a tool-use loop: it can list the sensors, request hourly summaries, drill into raw readings where something looks off, and pull the installation context for any room it's evaluating.

What it looks for is grounded in the physics of mold growth:

• Dew-point margin — how close surfaces are to the temperature where moisture condenses. An AC-cooled wall in a humid room can be condensing water invisibly even when the air "feels fine."
• Persistence — a humidity spike for an hour is harmless; 78% held for four hours a day, three days running, is a countdown. Duration above threshold matters more than peaks.
• Room context — the same numbers mean different things in a ventilated bathroom and a sealed closet. The analysis treats each zone individually.

One deliberate design decision: no seasonal adjustment. Mold grows on physics, not the calendar — relaxing thresholds "because it's summer" would simply mask real risk during the most dangerous months.

Step 4: Thresholds Tuned to Each Home

Alert thresholds work in two layers. Sensible global defaults cover every installation out of the box. On top of that, each location can carry its own overrides — and when we register a new room, an AI assistant proposes thresholds based on the regional climate and the room's description, which our installer can accept or adjust. A garage in coastal Florida and an air-conditioned bedroom should not be judged by the same numbers.

Step 5: Alerts Where You'll Actually See Them

When the analysis finds a risk, it raises a structured alarm — severity, a plain-language summary, and a recommended action — which is broadcast to subscribers via Telegram (with WhatsApp delivery on the way). No app to install, no dashboard to remember to check.

Subscribers can also talk back to the system: request the current status of every sensor, pull a 24-hour summary, or ask free-form questions ("should I worry about the laundry room?") that are answered by the same AI with full access to the home's data. Repeated alarms are rate-limited with cooldowns, so one bad weekend doesn't flood your phone.

The First Installation

The first production MoldSentrix system has been running in a home in Odessa, Florida — Tampa Bay's humid subtropical climate, where AC-cooled surfaces routinely flirt with the dew point. It has been exactly the proving ground the system needed: real weather, real AC cycles, real rooms with different personalities.

What's Next

The architecture was built to grow. On the roadmap: cumulative risk metrics (hours per week above humidity thresholds, hours of thin dew-point margin), year-over-year comparisons as history accumulates, WhatsApp as an additional alert channel, additional sensor types — CO₂, VOC, soil moisture — and a web dashboard for landlords managing multiple properties.

The core idea stays the same at any scale: measure continuously, analyze with context, and warn people while prevention is still cheap.