Analyze. Model. Deploy. Evaluate. Improve.
What we do
What we achieve
How we do it
We take advantage of the foundations of agile methodologies and proven methodologies for the development of machine learning systems, such as CRISP-DM, to build solutions tailored to your needs. We advise at every step of the lifecycle of the solution being implemented, with the goal of developing a product that is able to deliver the expected business benefits. This is our commitment: we only implement a solution if the preliminary analysis has proven that it will be able to reach the business objectives that have been established.
In order to minimize the risk of integrating a machine learning solution, we split the process into four phases:
Analysis of the situation
The analysis of the situation is the starting point of all our work. During this process, we evaluate the current state of the art, understand the expectations that your system must meet, and study the quality and veracity of the available data.
During this phase, we begin to understand how artificial intelligence and machine learning can improve the current situation of your company. In addition, our clients receive training on the practical application of machine learning within the company, so that the design of the solution being built is improved by the acquired knowledge. By doing this, we manage to establish a long-term alliance, in which our customers are increasingly benefiting from the solutions we develop.
Once the business requirements for the system to be developed are well understood, our team will build a first prototype, a proof of concept whose purpose is to validate whether the upcoming product will be able to deliver the expected benefits. This feasibility test usually consists of a rough prototype without any king of graphical interface, which is very far from constituting a final product, and whose objective is to validate if the current state of the art in artificial intelligence and machine learning is able to solve your problem. We study and process the data derived from the actual problem and then we obtain a set of preliminary metrics that indicate if the project will be able to deliver the expected benefits. With the result of these tests, we will prepare a first report with the technologies that have been tested and an analysis of what results, costs, and times can be expected in the following phase: the minimum viable prototype. If we foresee that the feasibility test is not promising, we will tell you before you invest more resources in a project that has little prospect of success.
It is important to consider one of the main axioms of machine learning and artificial intelligence: garbage in, garbage out. Meaning, that if your data is noisy, the results will also be noisy. This doesn’t mean that the data available are stored in a corrupted database or anything similar: it may be, for example, that there is not enough data to address the problem at hand.
Minimum viable prototype
Putting into operation a solution based on artificial intelligence or machine learning techniques is not risk-free: it can deliver important benefits to your business, but this is only possible if the appropriate resources are allocated, and your company should be well aware of this fact.
To minimize risks, the minimum viable prototype builds upon the results obtained in the feasibility test, but making use of the data and computational resources that the final solution would have. This implies that the results obtained from this prototype will be identical to those that would be obtained from the final product, but this prototype is probably not fully integrated into your company’s workflow. For example, things such as integration with your login system, a carefully crafted graphical interface, or integration with other computer systems of your company may be lacking. In addition, this prototype also allows us to establish the costs (in terms of both time and money) to develop the final product. We can see the minimum viable prototype as a preliminary solution to the proposed problem.
If everything went well in the previous phases and the reported results are promising, it’s time to build the final product. Here we will focus on all aspects that have been overlooked in the minimum viable prototype, putting all our knowledge in applications development.
Once the final product is finished and deployed, we will also be in charge of advising on its use, as well as solving potential problems that could arise later: we also offer continuous improvement, support, training, and full advice when using the implemented solution.
Pre-selection of employees