AI in leak detection only works once a system has seen the building it is protecting. Silviri is built around that reality. Every deployment moves through two phases — observation, then AI-enabled protection — and the AI is genuinely yours: trained on your building, tuned to your patterns, accountable to your operators.
Most leak detection vendors operate on fixed thresholds — a flow rate above X gallons per minute triggers an alert, a sensor in contact with water for Y seconds triggers a shutoff. These rules are simple to ship and easy to demonstrate in a controlled environment. They are also fundamentally limited, because every multifamily building has a different idea of "normal."
A 200-unit luxury high-rise generates a wildly different consumption signature than a 60-unit garden-style property. A pre-1980 building with galvanized plumbing flows differently than a 2015 PEX retrofit. A property with morning irrigation flows differently than one without. Static thresholds cannot accommodate this variation — they apply the same rule book to every building and produce the same predictable failures.
When thresholds are tight enough to catch real anomalies, they generate frequent false alarms during ordinary usage — morning shower hours, irrigation cycles, common-area appliance use. Maintenance teams begin ignoring alerts. By the time a real leak comes in, the alarm has lost its meaning.
When thresholds are loosened to reduce false alarms, slow leaks slip through. A toilet flapper running at 0.3 GPM costs the building thousands in monthly water charges and is invisible to a system tuned for catastrophic flow events.
A threshold tuned for a 200-unit high-rise is wrong for a 60-unit walkup. A threshold that works in summer fails in winter. Vendors cannot ship a single rule book that works across the diversity of multifamily buildings — so they ship a compromise that works poorly everywhere.
The fixed-threshold architecture forces a binary choice. The threshold can be set low enough to catch slow leaks, or high enough to suppress false alarms — but it cannot do both. Both directions of that choice produce expensive outcomes for the customer.
Silviri's architecture takes a different approach. Each building's idea of "normal" is learned from its own operating data. Anomalies are then scored against the building's own baseline rather than against a generic threshold designed for an average property that does not actually exist.
The result is not a vendor compromise. It is the difference between a fixed rule and a learned rule — and it lets Silviri achieve outcomes that fixed-threshold systems cannot reach simultaneously.
What looks like an anomaly in a generic threshold model may be perfectly normal in your building. Silviri knows the difference because it has learned your building's pattern. Operators receive fewer alerts, and the alerts they do receive carry meaning.
A continuous low-rate flow that looks like normal usage in a high-volume building is unmistakably anomalous in a small one — and Silviri scores it accordingly. Slow leaks, running fixtures, and degrading plumbing components surface as deviations from your baseline, not as signals lost inside a generic average.
Every Silviri deployment begins with conservative rule-based protection while the system observes your building. As the AI's understanding matures, its role in detection grows.
Every multifamily building has its own water personality — its own combination of resident habits, occupancy rhythm, daily and seasonal patterns. The combination is unique to your property, and it is exactly the combination that fixed-threshold systems cannot recognize.
Silviri's AI continuously tunes the system's behavior to the way your building actually operates. As your building changes, the model updates with it. What is normal at your building today may not be normal next year — and the system tracks that change instead of fighting it. The result is fewer false alarms during ordinary use, and faster, sharper detection when something genuinely goes wrong.
Each tier generates different data. The AI capabilities available to a deployment are bounded by the data the deployment produces — Detect-tier installations produce a thinner signal than Optimize-tier installations, and the learning curve reflects that.
Each capability is purpose-built for a specific operational problem in multifamily water management. AI is applied where it produces measurable improvement — not as a general label over the whole system.
We are happy to walk through how Silviri would apply to any specific property — including how the AI's role differs at the Detect, Protect, and Optimize tiers.