A few things I've built and worked on.
Performed a penetration test on a Metasploitable virtual machine, using Nmap, GVM, Metasploit, and Netcat to find and test security vulnerabilities. Successfully exploited three critical vulnerabilities to gain remote access, execute commands, and access sensitive system information. Documented additional security weaknesses and provided recommendations to improve the system's security.
Fine-tuned a FLAN-T5 language model on healthcare call-center conversations using masked response prediction. The model showed a major improvement in understanding context and generating responses that matched the purpose, tone, and structure of real agent responses, even when the wording was not an exact match. Achieved a 0.57 average ROUGE score compared to 0.05 for the untrained model.
Contributed to the early-stage development of an integrated machine learning workflow for non-invasively identifying and monitoring brown bears on the Kenai National Wildlife Refuge. Labeled keypoints on grizzly bear camera trap videos to build training data for a pose-estimation model, then trained YOLO on those keypoints to detect them automatically, which was used alongside tools like DeepLabCut-ReID and MiewID for individual re-identification. Helped lay the foundation the model would be built on in upcoming years, contributing to early discussions on study design and methodology as the project moved toward full-scale data collection.