As unmet mental health needs continue to rise, challenges in accessing and navigating resources and services remain a significant barrier to finding effective care. To address this challenge, MIRA, an AI-driven mental health virtual assistant, was developed [1], along with the MIRA library—a curated database of vetted mental health resources. This paper details the methods for designing an intelligent resource retrieval system that processes queries from a chatbot to identify relevant resources from our library. The proposed method combines term frequency and sentence embeddings to create vector representations of queries to perform similarity search for resource ranking. In addition, we incorporate query relaxation to leverage ontology-based hierarchical relationships. This ensures broader, yet contextually relevant search results. The proposed retrieval method outperforms traditional keyword-based and BERT-based methods in precision, as shown through our evaluation. The design and implementation of the ranking algorithm are detailed, including the construction of a knowledge graph and the integration of TF-IDF (Term Frequency-Inverse Document Frequency) based scoring. MIRA leverages the proposed retrieval method to rank resources effectively, recognizing user needs and connecting them to the most relevant support. By doing so, MIRA aims to bridge the gaps in traditional mental health care and improve accessibility.