Address

1410 Stanley Suite 606
Montreal, QC H3A 1P8


600 3rd Ave. Suite 222
NY, NY 10016, USA

 

Email

info@datacratic.com

Adresse

1410 rue Stanley 606
Montréal, QC H3A 1P8


600 3rd Ave. Suite 222
NY, NY 10016, USA

 

Courriel

info@datacratic.com

 

Location map

Contact Us

How can we contact you?
http://
CAPTCHA
This question is for testing whether you are a human visitor and to prevent automated spam submissions.
Image CAPTCHA
Enter the characters shown in the image.

Contact us

Datacratic Platform

At the heart of Datacratic’s technology is a machine learning platform that trains and executes predictive models. Supported by a complex structure of real-time data-flow engineering that ensures that the right data gets into the platform at the right time, the platform first introspects the structure of data and then extracts salient features for the predictive models.

The Datacratic Platform can separate signal from noise in real-time and then use the most valuable signals as feature vectors to train predictive models using multiple machine learning algorithms.

The platform pulls together various open source, published and proprietary machine learning algorithms. In many cases we have completely redesigned the algorithms with a focus on real world applicability. Real-time performance is paramount, we have engineered the feature vectors to operate extremely efficiently. Frequently we will replace an exact implementation of a slow algorithm with a fast, approximate implementation that provides 90% of the value but runs hundreds or thousands of times faster. The platform is surrounded by a simulation and training infrastructure that allows unbiased models to be trained and their results simulated against a real world dataset. 

The Datacratic Platform allows our products to be easily integrated into our clients’ platforms by providing I/O connectors into various data flows, while also ensuring data security. At Datacratic we develop elegant technical solutions that maximize the usefulness of real-time data flow. Our technology finds patterns that others miss, which automatically makes a big difference in the real world.

 

The Difference is Datacratic

Real-Time Updates

Datacratic’s platform is able to automatically take new data sources into account, which allows for real-time updates to the models whenever a new source of data is introduced.

What-if Scenario Testing

Our platform provides a rich and productive modeling environment that can be leveraged to provide what-if scenario testing and back-testing because modeling is performed via simulations that can be run with perfect fidelity.

Highly Compressed models

Data sources and models are highly compressed and stay on a single machine, similar to financial trading platforms.

Massive Scalability

The performance focused architecture of Datacratic’s platform is designed and engineered for massive scalability across cores and machines.

Economic Models

Economic models are used throughout the process, ensuring that data can be properly evaluated.

Online/Offline Learning Mode

Datacratic's predictive models are designed to operate in a hybrid online/offline learning mode:  online adaptation and periodic offline retraining.  Datacratic's platform is not a black box; it includes tools to visualise and understand what is driving the behavior of the models.

 
How it works:

 
Client Data
Datacratic does not have a proprietary source of data. The platform starts by ingesting client data either in batch or real-time.
Data Ingestion and ETL
Binary, json or xml; the platform can ingest data regardless of how it is being generated. The high-performance real-time endpoints can handle hundreds of thousands of events per second and scale by simply adding more servers.
Data Store
Data is stored in a format specifically designed for memory and CPU efficient machine learning. Feature generation has been designed right in the data containers so algorithms can be run on an extremely large quantity of data efficiently and easily. This creates a real-time path from raw noisy data to clean signal right in the datastore.
Model Generation
The first step in the machine learning pipeline is model generation where data is analyzed and features are extracted or rejected based on their value in predicting the final outcome.
Model Training
With a clean set of feature vectors the platform will train models without bias. Multiple models are trained in parallel, exploring different combinations of possible hyperparameters. The best performing model is then sent to production.
Simulation and Validation
All models generated by the platform go through cross-validation and simulation to protect against over-fitting and ensure that real world performance is similar to training performance. New data sources are introduced here and their value is determined before using them in production.
Final Model
The final output is an auto-generated, high-performance model which can be used to make predictions. The models are made available as a service to Datacratic's Products.
Production
Predictive models power the current suite of Datacratic’s products and they can also be applied in a number of other contexts. The platform has been designed for flexibility of integration into other systems and platforms.

Datacratic's predictive models are designed to operate in a hybrid online/offline learning mode: online adaptation and periodic offline retraining. Datacratic's platform is not a black box; it includes tools to visualise and understand what is driving the behavior of the models.