Visual Question Answering (VQA) is a rapidly advancing field within artificial intelligence, aimed at enabling machines to interpret and answer questions related to visual content. However, the majority of existing VQA models are developed for resource-rich languages like English, often overlooking low-resource languages such as Assamese. In this paper, we aim to bridge this gap by introducing an Assamese Visual Question Answering (AVQA) model. To the best of our knowledge, this is the first work in VQA for the Assamese language. The major contribution of this paper is the creation of the TDIUC-AVQA dataset, which involves translating the questions and answers from the original TDIUC (The Task Directed Image Understanding Challenge) dataset into Assamese. The translated content is then carefully reviewed and manually corrected to address any semantic and syntactic errors. Additionally, we also employ an Assamese-VQA model which uses a Bi-GRU as question encoder with self attention to effectively classify question-answer pairs. We perform both quantitative and qualitative evaluations of the AVQA model on the proposed TDIUC-AVQA dataset, offering valuable insights into its performance and relevance of VQA in low-resource languages like Assamese. Our AVQA model achieves 82.67% F1-score on the TDIUC-AVQA dataset. The sample copy of the dataset is available at https://github.com/Nazreena-Rahman/AVQA_CVIP2024/tree/main/TDIUC-AVQA-Dataset .