Our goal is continual research in the field of Artificial Intelligence.
The first pillar are our projects / orders, the second represents professional courses.

Emotional analysis

A trained, deep neural network recognizes feelings in texts
and assembles emotional image of personality.
The Emotional Graph across generations (left picture) is based on the analysis
of 2,000 handwritten letters of private correspondence.

PR: What is public opinion about _______?
HR: Analyzing Candidate Behavior - Is the New Candidate the right one?
Surveys: What do our customers think about our product?

Detector of Insults

As you may have noticed, we launched a DETECTOR OF INSULTS.
It is a typical application for classifying and analyzing text.
-> In this case, we've just focused on vulgarity.
The first version was based on the principles of deep neural networks, LTSM, embedding and word2vec.
The second generationhas been enhanced by a hybrid system that combines the power of Deep Learning with the ability of expert systems
to explain what word in sentence is vulgar.

Automated discussion forum moderator
Vulnerable call statistics in the call center

A simple mobile application was also released.

Project "keyword"

A smart network recognizes the said "keyword", for example, to launch a device.
The network has a 1D convolution, a spectrogram, two GRU layers and a dense NN network with a classic sigmoid.
NN has about a half million parameters. We tried to record our own voice commands and, surprisingly, understand our English. :-)
From now on, our networks listen to the word!

Image classification

"Training" the NN to make labels for photos is a little bit more complicated.
A overtrained CNN based on beloved InceptionV3 learned
the labels in just 10 minutes.
It's not perfect, but the results are breathtaking! 

- Google Images searching algorythm
- Blind people helping software

Photo analysis

During the studying advanced machine learning at the National Research University, the leading Russian university of economics, we let AI to recognize and tag images.

From the dataset of 50.000 images, the trained network most often confuses horses with dogs.  
The dogs confuses with cats, and when it comes to recognizing birds, NN thinks, It's the plane.
Isn't it magically human? :-)

Name generator

We tried to
use recurrent neural network  of 8.000 forenames
from different cultures to generate:

1) another similar names?
Result: Gawcalis, Heulis, Jesn, Asdee, Drarie, Nelap, Lean a Kantina

2) Forenames starts as "Jan" because it's such a beautiful name! :-)
Result: Janene, Janbie, Jann, Jankee, Janb, Jan, Jant

3) We have tried to generate card names of Magic
Result: „[U] Instant: Mad Aestaln“, „[R] Artifart: Elpe Phone“

4) .. and at the and we tried to generate Czech insults:
Result: kadik hoř, pojko vabo!, Lenokák, Čurdo plšička

Certificates / Education

When it comes to education we use: Coursera, Udemy and Udacity

We can strongly recommend courses:
Advanced Machine Learning Specialization (Coursera) 
Neural Networks and Deep Learning (Coursera)
Deep Learning for Business (Coursera)
Become A Robotics Software Engineer (Udacity)
Deep Learning (Udacity) – free
Expand Your Knowledge of Artificial Intelligence (Udacity)
Artificial Intelligence 2018: Build the Most Powerful AI (Udemy)
Elements of AI (University of Helsinky) – free

en_GB cs_CZ