João Pereira has completed an Integrated Masters degree in Computer Science from NOVA School of Science and Technology (17 average grade). In the scope of this degree, he was part of the project "CRM@DI", where he developed a CRM strategy that allowed different stakeholders from the Department of Computer Science at NOVA to better manage their interactions with enterprises regarding student's internships and joint masters/PhD thesis. During this project he was granted a research scholarship by NOVA LINCS. He has recently finished his masters thesis, entitled "Real-Time Human Action Localization In The Wild" under the supervision of prof. David Semedo (NOVA) and prof. João Neves (UBI), in partnership with the company DeepNeuronic (Neurónio Inédito) which funded the thesis in collaboration with the participating universities, through a research grant of the project DeepNeuronic PT2020 (CENTRO-01-0247-FEDER-113023). He was also involved with NOVA LINCS and SOCIA-LAB research groups. His work focused on developing optimization techniques that allow for the efficient deployment of Spatio-temporal Action Detection models in low computational resource scenarios, such as Video Surveillance related tasks. This thesis motivated a scientific contribution in the form of a paper submission at the IEEE International Conference on Multimedia and Expo (ICME2024). During his masters, he was an Erasmus student in Dresden, Germany, where he attended courses at TU Dresden for a semester. Currently, João is starting a PhD in Computer Science at NOVA, for which he was granted a studentship from Fundação para a Ciência e Tecnologia (FCT) for a PhD in a Non-Academic Environment (2023.03647.BDANA). He will conduct part of his research at NOVALINCS research group, and part (at least 1 year) at DeepNeuronic, in Covilhã, under the (co)supervision of the CEO Vasco Lopes. His current work is focused on introducing the task of Spatiotemporal Action Detection to Video Surveillance applications, leveraging the power of Large Vision-Language models to improve performance and to detect novel, previously unseen actions in Zero and Few-Shot settings. The output of this PhD will not only contribute towards advancing the state-of-the-art in Spatiotemporal Action Detection and Video Surveillance Understanding, but will also have a direct positive impact in DeepNeuronic's use cases (which provides Video Surveillance solutions using AI powered technology), as it will allow adapting a unique model to different client scenarios without the computational burden of generating a significant number of extra annotations and training the model from scratch. João's research interests include: Spatio-temporal Action Detection, Video Surveillance, Multimodal Video Understanding, Large Vision-Language models, Zero and Few-shot Learning and 3D Convolutional Neural Networks.