Imagine reading summaries of articles in any language, even if you don't speak them. That's the promise of cross-lingual summarization (CLS). However, building effective CLS models, especially for languages with limited data, is a tough challenge. Existing methods often struggle when high-quality, parallel datasets (original text and its summary in different languages) are scarce. This is where ConVerSum comes in. This innovative approach uses contrastive learning, a technique where the model learns by comparing and contrasting examples. Instead of needing massive parallel datasets, ConVerSum generates various candidate summaries in different languages from a single document. It then compares these summaries against reference summaries, learning to distinguish high-quality summaries from poor ones. This process involves a clever scoring system that considers both semantic similarity and language accuracy. ConVerSum is tested across a wide range of language pairs, even for low-resource languages like Bengali, Thai, Burmese, and Tigrinya, proving its effectiveness. When compared to large language models (LLMs) like GPT-3.5, GPT-4, and Gemini, ConVerSum shines, especially for languages where LLMs often stumble. It outperforms these powerful models in several low-resource language combinations, showing its potential for bridging the language gap. While challenges remain, such as computational requirements and the need for diverse candidate summaries, ConVerSum paves the way for more efficient and accurate cross-lingual summarization, bringing us closer to a future where language is no barrier to understanding.