When one feels unwell, it is crucial to arrange a time as soon as possible to meet a doctor for early detection of potential health-related problems. However, a relatively large number of Vietnamese people usually avoid going to the hospital as they are afraid of long waits at such crowded places, while the current COVID-19 pandemic means being at those places poses a higher risk of contracting the disease. For simpler health problems, people would prefer a solution that, given their symptoms, provides a reliable diagnosis in a shorter time. This study presents an approach in building a deep-learning-based disease predictor of health conditions conducted from given symptoms in Vietnamese. The proposed method combines a tokenizer and bi-directional recurrent neural networks and achieved an accuracy of 98.96% (compared to a certified doctor’s diagnosis) in selected test cases, demonstrating its promising capabilities in the task. The application is expected to easily be integrated into a mobile application and open the way for other deep-learning-based solutions which analyze people’s symptoms to help them have their health conditions diagnosed at home.