The remarkable performance of recent large language models has made the conception of new advanced tools to assist researchers in specialized fields feasible, provided these models are supplied with relevant specialized information. This article is the first in a series which aim to present our personal feedback in developing such tools within the context of studies on the bacterium Mycobacterium tuberculosis. One of the critical aspects of these developments involves extracting useful information from an extensive collection of over 100,000 research articles on this bacterium, which is highly specialized and diverse, and possesses unique characteristics. This initial article examines how information retrieval can be implemented in this context, based on our experience, and discusses optimal methods for recovering PDFs, extracting information, encoding it, and determining the appropriate type of vectorstore for storage. The approach presented here is not exclusive to this particular bacterium but can be extended to numerous other research areas.