Disseny i implementació d'un framework integral de MLOPs

Artificial intelligence and particularly some of its disciplines such as machine learning and deep learning have become key capabilities to solve many of the problems surrounding us or better the existing solutions. Consequently, many companies have invested and continue to invest large amounts of resources in incorporating these technologies into their daily lives. Within this ecosystem, although a few years ago it was rare to have people with knowledge of machine learning, today it is no longer the case. The new scenario faced by companies is not so much to develop models but to deploy these models from a laboratory phase to production environments. Although the development of machine learning is often associated with the produced models, the code itself represents a very small percentage of all that is needed to build a machine learning solution, so the management, maintenance, and deployment of models are today the biggest challenges of a company using artificial intelligence technologies. As a result, new tools are appearing to respond to these needs. Among these tools, end-to-end machine learning platforms tackle all the steps to bring a model from an experimental environment to the real world and once there, track and keep improving the deployed solutions. This master's thesis proposes the design of an architecture of a machine learning platform that can respond to the needs of the company in this area. To build this design, a series of previous steps are carried out such as an analysis of the context and needs of the company, as well as exploring existing solutions considered to be the state of art. Based on these studies, a design proposal is made which is then reasoned. The proposal is implemented and a series of experiments are carried out to deepen the functioning of the platform and ensure that it meets the requirements set. Finally, there is a discussion about possible alternatives. The completion of this thesis allows the construction of a platform with which to meet the company's daily challenges and improve existing processes. Both the product and the knowledge obtained represent the first iteration and may serve as a starting point for building a robust and flexible platform