This paper assesses populism in the 2022 Brazilian presidential elections, focusing on candidates' tweets. It employs supervised machine learning and active learning to fine-tune the BERTimbau algorithm. We train a model with two classifiers, one for people-centrism and another for anti-elitism. After an iterative improvement process by annotating high entropy observations, we run the model in a sample of ~10K tweets. Results indicate that while candidates tweet more frequently during campaigns, their populist rhetoric is more pronounced in the pre-campaign phase, with a notable moderation during the campaign and a sharp decline post-runoff.