Why Violet?

We are simple to work with, smart, fast and result driven local player with many years of experience. We love what we do and put quality and team-happines first.

Niklas Horn Lundberg
CEO, OptoSweden

”When creating our AI-interpreter we wanted to choose a team that understands the field very well and also have the abilities to really set up goals, beat them and overdeliver to any price. Violet has really proven that they both handle the machine learning part as well as the building of the infrastructure needed to have the best solution out there. It has been a great journey and results are over what we could imagine. With these guys in your team you will have bang for the buck top deliverance.”

Vasyl Garazd
CIO, Credit Express

”I believe that machine learning and scanning documents with machine learning using entity recognition is a very good thing for our company now and in the future. Working together with Péter Nagy and his team works very good and the openness that they have working with our team with knowledge transfer has been a great start for our machine learning competence to scale, with high knowledge base and proven results from various projects I can recommend the team to everyone who wants to start their AI-journey.”

Oscar Lidbeck
Business Innovation, Svea

”I have worked with Fredrik Schlyter on some various machine learning projects. To work with him and the crew has been very good cause what I have learned is that they come with both the creative aspect as well as the productive part, I can warmly recommend them to any team that wants and needs muscles on the machine learning side.”

JobsGet in touch

Here are some of the general fields to give you an idea.

Feel free to contact us directly so we can have a chat about this.

Computer vision

Object detection, image classification and segmentation.

Credit and fraud

Credit and fraud model development for the financial sector.

Natural language processing

Sentiment analysis, text classification and clustering.

Recommendation systems

Product and service recomandation for e-commerce and other industries.

Predictive maintenance & IOT

Time to event modeling and predictive maintenance for various industrial applications.

Time series forecasting

Estimation of future performance of time series data

Some of our latest projects

Here are some examples of how we have applied machine learning to our clients.


Credit Express Group

Saver Scandinavia


Here are the basic steps to succeed with projects.

Feel free to contact us directly so we can have a chat about this.

Workshop to understand and find

In the beginning of each collaboration we have a workshop where we try to find potential project candidates. During the workshop we give an introduction to the field of machine learning. We present projects that we have completed previously. A part of our offering is our unique insight into the current state of machine learning which we also talk about during the workshop. Once everyone has a better understanding of machine learning and the problems it can solve, we move on to identifying potential project candidates. It is very important that each project delivers business value for our customers, this is a common pitfall within the field of machine learning. We rank each project based on the business value it delivers compared to the complexities of development. After the ranking we select the projects with best business value versus complexity trade-off and start building.

Build a proof of concept to prove

Once we have settled for a project the next step is to build a proof of concept. First we do a rigorous data analysis in order to decide how to best build the product. We try to determine the volume of the data and the quality of the data. Size and quality of the dataset are two important parameters when building algorithms since this impacts which models we can use. It is very important that we decide the outcome of the project and how we measure the successfulness of what we build. The metrics that we use needs to have a tight coupling with the business value that we want to deliver. One metric could be the accuracy of the algorithm. In the early stage of the project we would decide on a minimum accuracy that is necessary to achieve some expected business value. In order to determine the full potential of the project we also attempt to calculate the maximum possible accuracy. Once we know which metrics to use and what the target is, then we can decide which method is best suited for this project. It is not always necessary to use machine learning, simple heuristics can also with depending on the complexity of the problem and the target accuracy that is needed to

Push to production to get it out

Now we have a proof of concept, but a proof of concept is only a small part of the whole solution. It is important to make the product accessible by pushing it to production. We have experience of pushing machine learning models to production. It can be done in a couple of different ways depending on the use case. We can deliver a python library which can be easily imported into existing projects. We can also deliver a docker container with a python REST API that can be accessed through HTTP. We also have experience of building a kubernetes infrastructure with RabbitMQ and the machine learning library as a queue consumer/producer. We also use CI/CD which works well with one of our core principles which is to try to reach a production ready state as early as possible in the project. The reason for this is to be able to deliver business value even if the accuracy of the models can be improved. Once we have reached a production ready state we can iterate quickly and continuously push out new model improvements.

Hand-over & knowledge transfer

The project is not complete until we have had a proper hand over. This can either be done at the end of the project or it can be done continuously during the project. We comment all the code that we write and we also make sure to have a test suite. We have learned that these components are essential in order to guarantee a good hand over. We are also able to offer training, if the organisation that is receiving the hand over lacks any knowledge that is necessary to use or develop the application further.

Here are some of the general fields to give you an idea.

Feel free to contact us directly so we can have a chat about this.

1. Data analysis

The first phase of any machine learning project is to analyze the data in order to find potential business value. We also evaluate the quantity and the quality of the data which is important when estimating different solutions. In order to maximize business value it is important to decide the scope of the solution and the acceptance criteria.

2. Clean data

The second step is to clean the data. Data is never perfect and model performance can be increased tremendously thanks to proper data cleaning. When working with text data that could mean removing stop words, when working with geospatial data it could mean geofencing bad coordinates, or removing duplicate records when working with transaction data.

3. Process data

Once the data is cleaned we need to prepare the data for our machine learning models. This means that we use mathematical transformations in order to create parameters that we feed to our models. The model can then use these parameters in order to understand how to solve the task at hand in an optimal way.

4. Train model

When the data has been cleaned and processed it is fed to our model which learns how to solve a certain problem from historical data. Everytime we receive new data, we can use this to train our model and increase its predictive power. We train several different models with different settings in order to find the model which is optimal for the task at hand.

5. Model revision

Once the models are done with the training we evaluate the models. The models are evaluated on data that they have never seen before in order to measure how well the models generalize to situations that they have never been in. The best performing models are then delivered to the production environment.

6. Production

Once the model has been trained and evaluated it is time to push the model to production. In most new machine learning projects it is important to quickly build a baseline model and then push it to production. Then the model will be able to deliver value quickly and it will also be possible to evaluate the model compared to existing solutions.

A data creative agency and solution partner focusing on AI and machine learning founded in 2017 in the heart of Stockholm.