<p>According to Ken Wyatt, the Minister for Indigenous Australians, "it is a fundamental right to speak your own language and use it to express your identity, your culture and your history". This critical ability is under threat, as indicated by the UNESCO Atlas of Languages in Danger, which reports that 40% of the world’s 6,700 spoken languages are on the brink of extinction. Despite the formidable challenges posed by a lack of written and spoken materials and inadequate preservation techniques, emerging technologies present promising avenues for preserving endangered languages. This research delves into the applicability of such technologies by focusing on two Colombian indigenous languages, serving as case studies. The aim is to trial various strategies in combination with the Transformer architecture and Transfer Learning to identify the most effective amalgamation for enhancing language translation and preservation. The study reveals that decreasing the number of encoding and decoding layers significantly boosts BLEU and chrF scores. Transfer Learning also demonstrates potential for improving model performance, even though improvements are not categorically guaranteed by its application alone. Notably, while increasing the dataset size typically improves results, the increment must be substantial so as to consistently elevate the model’s performance. Beyond its primary contributions, this research also resulted in the preliminary development of parallel corpora for two distinct endangered, low-resource Colombian languages.</p>