Solutions

We want to provide you with the freedom to build your own software

After three years of research and development to create innovative smart engines based on human cognition and psychology, we decided to release our work and give our customers the freedom to use it to create the foundation of their cognitive solutions.

This means that our customers will have following rights: the freedom to run our cognitive engines as they wish and for any purpose, the freedom to study the source code and adapt it so it does their computing as they wish and the freedom to redistribute copies of those engines, modified or not to who they want.

Providing to our customers this liberty is not only about giving them access to our source code. We think that this is not enough to ensure their freedom.

We made it easy for them to understand and adapt to their needs the source code. It is cleaned compiled, well documented and respecting programming norms and standards. They will have an installation and a user manual. These will help them to setup their cognitive solution quickly, either on a cloud server or on their premises. Based on our experience, this can be done in a couple of minutes.

We also made sure that our solutions are only based on non-proprietary software. They are ad-free with no additional premium features for which you would have to pay later nor any license price or pay-per-usage fee. We also grant our customers with the same rights to have access to future updates & upgrades.

Example : Feature chart for cognitive services (as of July 2019)

Without any kind of limitation

Our software solutions are build in C. We selected this language for good reasons. C is portable and the software can be compiled and used on many operating systems and runs very fast since the software, when executed, is not interpreted by an engine.

We also decided to use only ANSI C (C11) libraries to keep the source code as standard as possible. It is compiled with GCC and Make. Once you get the source code, only standard C libraries are needed. For each software, we also provide you with a project file generated by CodeBlocks, a free cross-platform IDE. With this file, it is easy for you to open and adapt our source code to your needs.

All our software products are accepting standard text files and will output either HTML files or XML files. None of those formats are encrypted.

All our solutions have been tested on a server running Ubuntu 18.04 LTS 64 bits. There is no other software product needed.

Pricing

We just have one price. Each engine will cost you 4 745€. For this price, you get the cognitive engine, the installation manual, the user manual and the source code. All future updates or upgrades are included.

We decided to keep it easy for customers to buy our solutions. You will get all you need to use our software and adapt it to your needs. In addition, since you do not pay on usage, since you have no limits, since there is no hidden feature or limitation, you can be sure that you will not have to pay more in the future. Our solutions can be hosted on a cloud server of your choice. In case you own a cloud solution, you can host our solution on your server and avoid paying extra costs to reserve another cloud resource or pay additional data transfer costs.

Estimated TCO for sentiment analysis

Sentiment Analysis

Sentiment analysis is all about processing natural language to assess and identify affective states. Alex, our cognitive engine, will take care of this.

Our sentiment analysis solution is using a knowledge-based technique which will classify text by affect categories (positive or negative). We have assigned to each of these categories a list of related unambiguous words which have been taken from the latest clinical psychological researches. We added to each of these words their related semantic field. The result is available within a specific VirtualBrain. The current measured recognition rate is 94%.

The English version has 602 categories counting 2279 different words. The French version has 498 categories counting 2772 different words.

Sentiment analysis can be applied on large collections of texts such as voice of the customer material, web pages, online news, internet discussion groups, online reviews or survey responses, blogs or social media content.

Performance testing : we have assessed a verbatim (test file contains 181 words (889 characters, 894 bytes)) 40 000 times using Alex. This computing was done in 19 minutes and 50 seconds, using 4 cores on an HP Server (Intel® Core™ i7-5500U CPU @ 2.40GHz) running Ubuntu 18.04 LTS 64 bits. Anna is able to assess on this infrastructure 33,61 verbatim per second, representing a total of 87 126 050 assessment per month.

Emotions assessment

Emotion is a mental state which can be seen as the result of cognitive process. It is a positive or negative experience, producing different physiological, behavioral and cognitive changes. The role of emotions is to motivate adaptive behaviors. Anna, our cognitive engine, is able to identify these.

Our solution, able to assess emotions, is using a knowledge-based technique in order to identify and classify eight primary emotions which are joy, sadness, anger, fear, trust, disgust, surprise and anticipation. Those primary emotions are then linked together, are influencing each other, and those associations will form a set of more complex emotions. In total, 32 emotions can be identified. Those emotions can then be used as an essential input to other cognitive processes such as reasoning, decision making or problem solving, to adapt, alter or influence the response of the whole system.

Emotional assessment can be applied on large collections of texts such as voice of the customer material, web pages, online news, internet discussion groups, online reviews or survey responses, blogs or social media content.

Performance testing : we have assessed a verbatim (test file contains 181 words (889 characters, 894 bytes)) 40 000 times using Anna. This computing was done in 11 minutes and 41 seconds, using 4 cores on an HP Server (Intel® Core™ i7-5500U CPU @ 2.40GHz) running Ubuntu 18.04 LTS 64 bits. Anna is able to assess on this infrastructure 57,06 verbatim per second, representing a total of 147 902 996 assessment per month.

Human needs assessment

Everyone has a set of universal basic needs (about 40), with individual differences on these needs leading to the uniqueness of personality through varying dispositional tendencies for each need, creating an internal state of disequilibrium; the individual is then driven to engage in some sort of behavior to reduce the tension. This is the moment when needs become active and this is what Lisa is identifying.

We created Lisa, an artificial intelligence, for this purpose: be able to understand in “real-time” a given situation and find the set of active needs to be satisfied.

Better understanding customer needs is a key thing for each business. And it takes time to ascertain customer’s emotional and material needs. What would be the benefits for your business if you could analyze conversations, business emails, surveys or feedback’s and get an understanding about what’s happening and about what customer needs?

Primary and secondary human needs

And knowing these needs will make it easier to provide to people what they are looking for and adapt the actions taken by the machine to the current context.

Human need assessment can be applied on large collections of texts such as voice of the customer material, web pages, online news, internet discussion groups, online reviews or survey responses, blogs or social media content. The current measured recognition rate is 91%.

Heatmap produced with the result of an assessment made for an Amazon.com use case (Feb. 2019)

Performance testing : we have assessed a verbatim (test file contains 181 words (889 characters, 894 bytes)) 40 000 times using Lisa. This computing was done in 16 minutes and 12 seconds, using 4 cores on an HP Server (Intel® Core™ i7-5500U CPU @ 2.40GHz) running Ubuntu 18.04 LTS 64 bits. Lisa is able to assess on this infrastructure 41,22 verbatim per second, representing a total of 106 666 667 assessment per month.

Understanding

Understanding is a cognitive process and can be defined as a procedure of attaining knowledge about oneself or other people or of understanding the meaning or significance of something, like a term, idea, argument, or occurrence. Understanding is correlated with the ability to make inferences (reasoning). While representations are replacing physical objects by their corresponding symbolic representation (the prototype) to enable cognition, understanding will use those symbolic representations to find out a general meaning. This is what Lucy, our cognitive engine, will do.

We provided this ability to understand to Lucy, a cognitive engine. We based our approach on how information is acquired by our senses and processed by our neurons, on how information is used by mental activities or cognitive processes to be able to understand a given situation.

This assessment can be applied on large collections of texts such as voice of the customer material, web pages, online news, internet discussion groups, online reviews or survey responses, blogs or social media content. Lucy will understand the content and provide you with a list of keywords representing the general meaning of your unstructured content, creating new knowledge and learn in real time.

Assessment of online reviews in French for a OnePlus Smartphone.

Meaning recognition / use case specific cognitive engine

Humans retain information in their long term memory. This long term memory (our database) is organized in a structured way. The information is stored and linked to other information to form a complex network. These links will enable the transmission of some characteristics of one information to another information so that both will share a set of common properties. This can be considered as a semantic network where many information are linked together by a set of common properties conveying meaning. Allan, our cognitive engine, is able to work with this kind of categorized structure to recognize the meaning contained within unstructured data.

This structure is made of symbolic classes. You have 3 types of classes: the object, his prototype and his abstraction (example: Woodpecker, Bird, Animal). Each class contains information stored in properties such as functional properties and descriptive properties, enriched with their related semantic fields. For a given class, his hyponym is a child class which does inherit of all the properties of the parent class. For a given class, a superordinate is a parent class which holds all the common properties. Once stored (in the long-term memory) in categories, knowledge can be represented, can be used when reasoning, making decisions and solving problems.

We have called this categorized structure the VirtualBrain, a data file containing all the information, all the properties and all the needed relations between these objects classes defining your business context.

When a VirtualBrain is prepared, descriptive properties are enriched with their respective synonyms. We then need to clean this structure to remove those synonyms which are not conveying the expected meaning. At the end, the specific VirtualBrain is holding all the necessary information to recognize the particular meaning that we are searching for.

Some examples of specific VirtualBrain that we have prepared for our customers: be able to identify people requesting support, identify people looking at the competition, identify people requesting a duplicate of a document, identify people complaining, identify people requesting a meeting, etc… All these specific use cases are using a dedicated VirtualBrain prepared to identify the meaning contained within unstructured text.

Performance testing : we have assessed a verbatim (test e-mail containing 24 words (133 characters, 146 bytes)) 40 000 times using Allan with a specific VirtualBrain prepared for one of our Customer. The purpose was to assess the content of this e-mail to detect if the content is related to someone asking for a meeting. This computing was done in 3 minutes and 37 seconds, using 4 cores on an HP Server (Intel® Core™ i7-5500U CPU @ 2.40GHz) running Ubuntu 18.04 LTS 64 bits. Alex is able to assess on this infrastructure 184,33 verbatim per second, representing a total of 477 788 018 assessment per month.

Object categorization workbench

Cognition is enabled by using a specific categorized data structure that we called the VirtualBrain. Jack is our dedicated workbench that is used to prepare this data structure based on the context related semantic network of your use case.

Jack is a workbench dedicated to create VirtualBrains. It can be used to describe the objects belonging to your semantic network and how these are linked. For each object, you can specify descriptive, functional and dynamical properties. These properties can be used to enable cognitive processes such as reasoning, decision making or problem solving. If needed, descriptive properties can be enriched with their synonyms so that more knowledge is added to your specific VirtualBrain.

Our VirtualBrains are holding all the necessary information to enable cognition and core mental activities such as learning and forgetting.

Once created, VirtualBrains can be used by Alex, our cognition engine dedicated to recognize meanings or can be used as a cognitive data structure for your specific engine.

Our testing procedure

Testing cognitive software is not just an IT activity since it does involve specific psychological knowledge. At Beamak, testing is only performed by cognitive psychologist which is a guarantee that the result of the assessment performed is reliable.

The data set that we use to test our cognitive computing solutions has been assessed by a cognitive psychologist who gave his conclusion about the psychological state that can be found. This manual process time consuming since each data needs to be analyzed with great care. Results are then reported and only then the data set is processed by our cognitive solutions. The outcome of this assessment is then compared to the one found by the psychologist. If both are equal, the assessment performed by the cognitive engine can be considered as valid. When we report a recognition rate, it is based on this assessment.

A photographic picture of the documents used by the psychologist to check the assessment made by the cognitive engine