This paper presents a hybrid recommendation system for identifying similar environmental sustainability projects using a combination of deep learning and text-based similarity measures. Our approach preprocesses datasets by encoding numerical attributes and leveraging TF-IDF vectorization for textual descriptions. An autoencoder reduces the dimensionality while preserving essential feature representations, enabling the computation of cosine similarity between projects. The evaluation framework uses Precision@K and NDCG@K to assess the accuracy of recommendations across multiple datasets. The experimental results demonstrate the feasibility of this hybrid methodology in identifying relevant project similarities within environmental sustainability initiatives. Our findings suggest that integrating deep learning-based embeddings with traditional text processing enhances the identification of related projects, providing a valuable tool for sustainability analysts.