Designed and implemented a 𝗺𝘂𝗹𝘁𝗶-𝗺𝗼𝗱𝗲𝗹 𝗔𝗜 𝗱𝗲𝗳𝗲𝗻𝗰𝗲 𝘀𝘂𝗶𝘁𝗲 covering the full spectrum of AI-powered cybersecurity, structured as four end-to-end research modules with individual repositories linked as Git submodules. 1) 𝗡𝗲𝘁𝘄𝗼𝗿𝗸 𝗙𝗹𝗼𝘄 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 (Deep Learning) Built a deep learning pipeline on the CICIDS2017 dataset to classify 𝗯𝗲𝗻𝗶𝗴𝗻 𝘃𝘀 𝗺𝗮𝗹𝗶𝗰𝗶𝗼𝘂𝘀 𝗻𝗲𝘁𝘄𝗼𝗿𝗸 𝗳𝗹𝗼𝘄𝘀. Investigated class imbalance with weighted loss functions, conducted 𝗳𝗲𝗮𝘁𝘂𝗿𝗲-𝗯𝗶𝗮𝘀 𝗮𝗻𝗮𝗹𝘆𝘀𝗶𝘀 (destination port shortcuts in BruteForce attacks), and performed hyperparameter sweeps across FFNN architectures with SGD, SGD+momentum, and AdamW optimisers. 2)𝗠𝗮𝗹𝘄𝗮𝗿𝗲 𝗔𝗣𝗜-𝗖𝗮𝗹𝗹 𝗖𝗹𝗮𝘀𝘀𝗶𝗳𝗶𝗰𝗮𝘁𝗶𝗼𝗻 Classified malware families from dynamic API-call execution traces using 𝗙𝗙𝗡𝗡𝘀, 𝗥𝗡𝗡𝘀, 𝗟𝗦𝗧𝗠𝘀, and Graph Neural Networks (𝗚𝗿𝗮𝗽𝗵𝗦𝗔𝗚𝗘, 𝗚𝗖𝗡). Represented execution traces as directed graphs to extract structural behavioural features. Achieved 𝟵𝟴.𝟭% classification accuracy. 3) 𝗨𝗻𝘀𝘂𝗽𝗲𝗿𝘃𝗶𝘀𝗲𝗱 𝗡𝗲𝘁𝘄𝗼𝗿𝗸 𝗜𝗻𝘁𝗿𝘂𝘀𝗶𝗼𝗻 𝗗𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 Detected zero-day network attacks with no labelled attack data, using One-Class SVM, 𝗔𝘂𝘁𝗼𝗲𝗻𝗰𝗼𝗱𝗲𝗿-𝗯𝗮𝘀𝗲𝗱 𝗮𝗻𝗼𝗺𝗮𝗹𝘆 𝗱𝗲𝘁𝗲𝗰𝘁𝗶𝗼𝗻 (reconstruction error), DBSCAN/K-Means clustering, and t-SNE/PCA for visualisation. Achieved 95% precision on the NSL-KDD dataset. 4) 𝗡𝗟𝗣 𝗳𝗼𝗿 𝗠𝗜𝗧𝗥𝗘 𝗔𝗧𝗧&𝗖𝗞 𝗧𝗮𝗰𝘁𝗶𝗰 𝗥𝗲𝗰𝗼𝗴𝗻𝗶𝘁𝗶𝗼𝗻 Classified real-world 𝗯𝗮𝘀𝗵 𝗰𝗼𝗺𝗺𝗮𝗻𝗱 𝘀𝗲𝘀𝘀𝗶𝗼𝗻𝘀 into MITRE ATT&CK tactics (Discovery, Execution, Persistence, Defence Evasion) using TF-IDF, Word2Vec, Bidirectional LSTMs/GRUs, Transformers (BERT, UniXcoder) 𝗥𝗲𝗽𝗼 𝗨𝗥𝗟: https://github.com/RenatoMignone/AI-Driven-Threat-Detection-Research