brain-ai-platform.svg

Unrivaled machine learning technology

Developed to specifically address data modeling and prediction challenges across retail, the Celect Prediction & Optimization Platform is composed of proprietary innovations coupled with best-in-class open source components such as Docker, Spark, Kubernetes, and Airflow. 

With this unique framework, it is able to rapidly ingest large amounts of data, model the data, and expose the resulting predictive data through a set of flexible API’s for bulk and ad-hoc querying. 

RETAIL-PREDICTION-OPTIMIZATION-PLATFORM@4x.png

THE KEY COMPONENTS

Similar to a schema-less database, the Celect Prediction & Optimization Platform provides general-purpose data storage and out-of-the-box querying functionalities. To provide the best data-driven decision support to companies across industries, it acts as a “prediction database” by offering predictive queries solving many of the most complex problems while leveraging regression, classification, time-series, matrix, and tensor completion. 

The Celect Prediction & Optimization Platform − composed of the Management Framework, Celect Engine, Data Layer, and UI Framework makes extremely accurate predictions, while building and optimizing models against constraints, across structured and unstructured data types. This power and effectiveness has been proven across Fortune 500 retailers.

data-model-blue-1.png

Ingest & Transform

The Celect Prediction & Optimization Platform consumes large amounts of data on-demand or on a schedule from various disparate sources to build a data store necessary to answer complex questions. 

Model 

The Celect Engine is the workhorse of the platform - built with computational speed and scale top of mind. As soon as transformed data enters the Engine, it begins building an exponential amount of predictive models from a library of algorithms. At any given time, the Celect Engine can process hundreds of millions of data points for analysis. 

Query & Visualize

The front-end UI and available APIs enable real-time querying, visualization, and reporting. As part of the Celect Prediction & Optimization Platform, retailers can leverage purpose-built solutions for critical insights and decision support for their business.  

We are able to answer common, yet difficult questions by modeling data in a unique way. The question often includes an optimization challenge – so not just predicting, but balancing a number of objectives at once. The desired result are answers to an "optimized" set of objectives considering your existing business constraints, all while balancing sparse, ever-changing data with extreme accuracy. This accuracy enables Fortune 500 retailers to boost revenues and margins by millions of dollars.

EMBRACE SPARSE DATA

Most companies have an abundance of data strewn throughout various systems, but the data on each customer is much smaller than the total. This leaves holes, which require “stitching” together these missing areas by combining data and other descriptors to fill these holes.

ACCURACY IS CRITICALLY IMPORTANT

As new data is consumed from more sources, further context is established and model accuracy exponentially improves. The result is extremely accurate predictive responses to queries across your business.

SPEED OF DATA CHANGE

Data is never stagnant, and constantly changing. In many industries and applications, predicting trends is extremely valuable. This is a challenge as new data becomes available and changes in existing data occur— e.g. transactions, user browsing patterns, or new products. The Celect platform transparently updates its prediction models in the background as new trends and patterns are captured.

BUILT TO CONSUME VARIOUS TYPES OF DATA

In many industries, data exists in various formats such as categorical data (e.g. department and product category), text data (e.g. product description), or image data (e.g. images of products). The Celect platform seamlessly converts both structured and unstructured data types to the appropriate internal representations without any user intervention to drive accurate and fast predictions.

patent-icon-light.svg
 
Behavioral Relationship Completion (1)
patent-icon-light.svg

Behavioral Relationship Completion (2)
patent-icon-light.svg

Prediction dB/ Completed Results w Confidence
patent-icon-light.svg

Multiple Objective Output
patent-icon-light.svg

Sparse Data & Missing Value Completion
patent-icon-light.svg

Data Computation above Memory Capacity
"Demonstrated that stitching pair-wise rankings together...is a more accurate representation of customer sentiment than relying on...a typical five-star scale."

"Algorithm predicted car buyers' preferences with 20 percent greater accuracy than existing algorithms."

"An algorithm that can predict trending topics on Twitter...up to four or five hours in advance."

"A new algorithm that can, with 95 percent accuracy, predict which topics will trend an average of an hour and a half before Twitter's algorithm."

data-funnel 600x400.jpg

prediction Database

A general overview of the pDB - a scalable, flexible, predictive platform using heterogeneous, multimodal data  

data-funnel 600x400.jpg

prediction Database (pDB)

A technical overview of pDB queries, schemas, and prediction types 

IQT-Article-Devavrat-Fall2015-300x200.png

Predictions with Big Data

The state of big data and the requirements to efficiently provide predictions of the unknown.