Named Entity Recognition (NER) is an important process for the treatment and analysis of textual documents, in different languages and specific domains, to solve tasks such as translation, automated summarization, and conversation, among others. In this paper, we propose a machine learning model for identifying fine-grained entities in the Spanish language and for the legal domain. The model is a bidirectional recurrent neural network that uses LSTM modules (Bi-LSTM) and is trained using a dataset built with Colombian legal domain contracts and the entities are identified and labeled by experts. The model has been thoroughly validated and tested, showing outstanding results in entity recognition, despite the fact that there are not many corpora related to contracts in Colombia. Our research findings indicate that the legal context of a country influences information retrieval about legal contracts.