Adaptix augments human judgement and business expertise with sophisticated technological capabilities to deliver the most impactful and valuable information, when it’s needed most.

Continuous learning

Adaptix’s engine builds its model of the input time series “on the go”, as the data comes in. The system never stops learning - unless it is commanded to do so. If a sudden change of context occurs, Adaptix will first flag an abnormal behaviour but will progressively adapt to this new context, as it “learns” its new behavioral pattern.

Unsupervised, adaptive modelling

Adaptix does not require user input, such as model selection, model parameters or meta-information (e.g. summary statistics or seasonality) about the dataset to start building its model. Little to no configuration and domain expertise (such a data science know how in the case of Machine Learning models) is required to get it up and running.


Sensewaves’ technology can be applied to single (univariate) data streams as well as multiple (multivariate) data streams. The latter can be used to analyse the joint behaviour of several variables together, which in Adaptix terminology are called “stream groups”. More generally, one can use groups to analyse several variables that are expected to have correlated behaviours.

AI Transparency & Explicability

Unlike the majority of machine learning systems where the reasoning mechanism is a «black box » to the human operator, Adaptix keeps all intermediate steps of the model transparent and has the ability to provide justification behind its results. This “White box” characteristic of Adaptix provides human operators with the tools to evaluate the computer logic behind the results, double check their validity and, if necessary, tweak the system’s analysis mechanisms.

Speed & Performance

There is a known tradeoff between “batch” and “streaming” data analytics systems, when it comes to performance. Streaming usually allows faster response times but will limit the data that can be analyzed to a subset or an approximation of it. Batch analysis on the other end, will allow the analysis to consider the whole historical depth, but this might imply long running computations. Adaptix blurs the line between both methodologies, achieving query response times of a few milliseconds to a few seconds regardless of the volume of data.


Building the knowledge base inside Adaptix


Adaptix is based on a proprietary graph-like representation technology of time-series data called SDMG (Stream Deep Memory Graph). Each such graph represents a projection of the time series in a feature/time resolution space. The initial data stream can thus be represented by a collection of interconnected SDMGs.

The representation through SDMG is the core mechanism for time-series modelling inside Adaptix. SDMG represent the data stream in time and space dimensions. Using parallel SDMGs of different time/space resolutions, Adaptix may provide associations of the present with similar experiences from the past in a blink of an eye.

Through SDMG, learning is continuous and adaptive, whereas reasoning becomes transparent. Operations such as forecasting, pattern search, anomaly detection, classification become efficient and quick, even for the most complex, heavy and fast data.


Sensewaves’ VirtualMDA is the technology behind Adaptix.Grid load state estimator. VirtualMDA minimises costly and time-consuming installation of sensors and systems by grid operators. Through novel graph and electrical modelling algorithms, Adaptix.Grid works with the sensors you already have to create virtual meters at non monitored points.

As a result, you obtain a complete grid model capable of instantly providing load/voltage flow data covering 100% of your grid (transformers, feeders, switchgear, cables, cable boxes, etc.) even if a fraction of the network is actually covered in hardware metering equipment / smart-meters.

Applications with DSO clients have demonstrated that speed of analysis remains high and differentiates Adaptix from other analytics tools on the market, while observing a virtual metering accuracy performance of 95% on average.