Validated Infrastructure Estimation of Risks, Events and Signals

Your infrastructure already produces the data. We turn it into foresight.

We read your installation's telemetry, validate every value and publish forecasts with their reliability, so that whoever decides knows how far they can rely on them. We always start read-only, without touching the control of your installation.

Request a data assessment See real examples

Every forecast is published with its confidence band over the process quality zones, and only up to the horizon where it is reliable.

Built and validated on real data from a working wastewater plant and an instrumented monument, with the method published and under review in scientific journals.

What we do

From raw data to decisions made in time

Validated data

We detect gaps, drifts and impossible values before they reach any calculation, and correct them only when there is a basis to do so. A long outage is recorded as a documented gap, never filled with invented data.

Forecasts you can rely on

We compare several model families for each signal and horizon, quantify the uncertainty and discard physically impossible results. Each forecast is published with its confidence and only as far as it can be defended.

Risks and events in time

Clear indicators with quality ranges, warnings when data stop arriving or a signal approaches a limit, and alerts early enough to act. Your team decides, with more information and more time.

Instrumented infrastructure shares the same problems of integration, temporal quality and time-series use, so much of the analytical core is common to a wastewater plant, a monument or an industrial facility. Input adapters, physical rules, indicators and thresholds are specialised for each domain with expert judgement.

Examples

What the operators see

Real screenshots of the platforms in operation, with each installation's real data.

3D model of the wastewater plant with the sensors of each zone and the status of their variables — **3D model of the plant** Each zone of the model shows its sensors, the latest value and whether it is at risk, out of range or without data.

Reactor temperature: observed series and forecast for the coming hours — **Forecast with its horizon** Observed data and forecast are shown together, and only up to the horizon where the model has proven reliable.

Wastewater plant ammonium with the anomalous values marked on the series — **Anomalies flagged, not hidden** Impossible values and frozen-sensor stretches are flagged and set aside before anything is calculated.

Confidence table of the monument forecasts by variable, horizon and model — **Confidence explained** For each variable the platform states how far the forecast can be trusted, which model produces it and why some variables have none.

Real cases

Two different worlds, the same engine

Web application of the Cheles wastewater plant: plant status, indicators and access

Water · Cheles wastewater plant, Badajoz

A working plant, not a laboratory pilot

Built and validated on a plant in operation managed by PROMEDIO, with irregular sensors, data gaps and biological processes. High-accuracy forecasts for key variables, documented sensor health and a real instrument failure handled without inventing a single value.

Real temporal validation of structural convergence Z2 at Portezuelo: observed signal and five models compared over five days

Heritage · Portezuelo Castle, Cáceres

The same core, on a monument

The structural monitoring of the castle, with its cracks, tilts and environmental conditions, runs on the same core through input adapters and heritage-domain rules, without duplicating the platform.

Method

Rigour that can be audited

What can be examined is not the claim that a model is good but the protocol that selected it. The served model is chosen by comparing several families against a demanding baseline, over many time cuts that mimic real use, penalising flat forecasts that look accurate without being so. Uncertainty is calibrated so that the confidence band reflects the real probability.

The method is published and under review in scientific journals, and any technical counterpart can audit it on request. Honesty with the data is part of the same contract: if a sensor fails, the system says so and sets it aside instead of hiding it.

Anomaly detection on ammonium: original signal over plausibility zones and the detectors compared — Real detection on the pilot plant's ammonium: the selected detector flags physically impossible spikes and frozen-sensor stretches.

How we work

We start small and grow with the evidence

Data assessment

An audit of your sources, quality and coverage, with a clear result on what can and cannot be anticipated with the data you have. If your data cannot support forecasts, we tell you before charging for a deployment.
Commissioning

Integration with what your installation already has, with no works and no replacement of sensors or supervision systems, and configuration of the indicators and forecasts specific to your asset.
Continuous service

Role-based web access, continuous forecasting, alerts, support and periodic model review, because an unattended model degrades.

We also work for suppliers in the sector, optimising their algorithms and adding forecasting that integrates into the platform they already sell.

Team and backing

University, a real plant and an ecosystem

We are a team from the University of Extremadura combining computer engineering and artificial intelligence with construction and building expertise. The project grew from years of work with a plant in operation and an instrumented monument, with scientific publications backing the method, and is supported by FUNDECYT-PCTEx in creating the company.

Let's talk about your data

If you run an instrumented installation that only looks backwards, the initial assessment tells you, with your own data, what can be anticipated. It is the lowest-commitment way to start.

Write to the team

The team replies to you directly. jamateosb@gmail.com