Before starting a machine learning project, we need a clear and well-defined goal. It is necessary to define the problem we want to solve as well as its business value. In addition, we check whether we have all the data needed to solve the specified problem, how this data is delivered and if it should be made anonymous. In the rapidly evolving world of AI, it is also important to research whether we can build upon previous academic research.
Data is the fuel of artificial intelligence, which is why we need to know your data, inside and out. We search for the best machine-readable representation for our machine learning model, after which we use known statistic methods to normalize data, detect outliers and fill in missing values. Furthermore, we document our findings in a report to improve the quality of your data. If necessary, we also annotate the data for the model building step.
While staying up to date with the latest developments, we carefully select suitable algorithms to create an efficient and accurate model. Each algorithm has so-called hyperparameters for the selection, which results in an iterative process to align them and evaluate the results. Our experiences during this process are summarized in a report, in which we compare and describe all methods and hyperparameters.
experience & interface
Because of the ever-evolving nature of data, any worthwhile AI model needs human validation for periodic training. And of course, you want to know exactly how well your new top employee is performing. Therefore, we use the latest technologies to create interfaces that bridge the gap between human and machine. Based on the feedback of your domain experts, our interfaces evolve to make sure that the end result is efficient to use.
Integration and deployment
As soon as development is ready, our brains need a jar or place to live and develop. In this final phase, we ensure that our application is completely ready to run independently. We provide the integration of the application with the customer's system and ensure its portability and scalability using technologies like Docker and Kubernetes. Finally, we deploy your application either on-premise or on any of the major cloud platforms.
Both AI's and HI's (Human Intelligences) make mistakes, which is why we believe AI systems should team up with humans.
When AI should take care of annoying and repetitive work, an employee can be promoted to AI Manager to check and adjust the AI work where needed. This allows our applications to work outside the lab environment, so we can succeed in improving and adapting them, in the long run, to last for years. It remains important for people to continuously train the algorithm and, if necessary, also validate it.
It is certainly not our intention to replace people but to shift their focus to value-creating activities and hand over unpleasant activities to AI.
AI is more than just AI
A good AI application is more than just a model. With our end-to-end approach, we take care of the entire process: starting from an idea or brainwork, we proceed to pre-process. We focus on the data pipeline after which we move on to building the model, front-end, and back-end design. The final step is to complete the project via deployment in the cloud.
An approach that made us the ideal partner for the so-called "Cold Start" scenarios, in which we design an AI application without any available data. We build AI models that are fully trained by the end-users. Initially, this training is quite intensive, but over time the AI model requires less correction until it can function almost completely independently.
The right tools for the job
We believe that each project requires a unique approach. Based on the needs of the customer, we choose the best possible solutions for each project and/ or problem.
The philosophy of "use when possible, build when necessary" is central to this approach. We use pre-built services or products where possible to reduce costs and development time as much as we can.
Only when the use case requires a customized solution, we use our extensive domain knowledge and experience with frameworks such as Tensorflow to build powerful custom AI models.