This article presents the recommendation system of Personas, a microservice-based platform designed to assist Human Resources (HR) teams in streamlining the recommendation and presentation of candidates to clients based on posted project descriptions. Personas offers functionalities for recommendation, automatic generation of tailored curricula and motivation letters, and conversational support through client- and consultant-facing chatbots. At its core, the recommendation system suggests relevant projects posted by clients to each candidate on a daily basis. It leverages both structured and unstructured textual data, including web-scraped content, user-uploaded documents, curated profiles, and texts generated by Large Language Models (LLMs). All documents are embedded into a shared semantic vector space, enabling fast similarity computations and facilitating Retrieval-Augmented Generation (RAG) workflows. The recommendation pipeline consists of a two-stage process. First, lightweight pre-selection models apply filters and semantic similarity metrics to narrow down the pool of potential assignments. Then, in-depth analyses using LLMs provide refined compatibility assessments. Notably, the LLM-based evaluations serve not only to improve ranking precision, but also as high-quality proxy labels for evaluating and improving pre-selection models. This paper describes each stage of the pipeline-document collection, structuring, curation, pre-selection, and LLM-based analysis-and presents quantitative results demonstrating the system's effectiveness on a large-scale dataset.