Today’s Tunisian real estate market platforms do not provide sufficient information for those seeking to make well-informed property buying decisions. Moreover, retrieving this information is challenging, as it exists in different places and is presented and structured in heterogeneous ways. This data should be cleaned, stored, and modeled in a flexible data structure, enabling the centralization of relevant real estate offers. The use of classical machine learning (ML) methods presents several challenges, particularly in terms of data preparation are complex. An application was developed using ML algorithms for data preprocessing, classification, feature extraction, and price estimation, achieving a classification accuracy of 85% and a recommendation precision of 0.79 using cosine similarity over TF-IDF. In addition, this article explores the potential impact of incorporating Large Language Models (LLMs) to further refine text analysis, particularly in handling the Tunisian dialect and automating data preprocessing. While LLMs were not yet integrated into the current implementation, their prospective use is discussed as a future direction to improve the interpretation of listing descriptions, extract key features, and reduce manual intervention. This work provides a practical contribution to real estate analysis in Tunisia and outlines perspectives for enhancing future systems with advanced language technologies.