I am a computer science graduate from Boston University interested in artificial intelligence, security research, and software engineering.
My most recent role was as an AI/ML team lead at BU Spark!, an experiential learning lab and quasi tech consulting firm with clients including the Harvard Herbarium and a handful of social impact nonprofits. In my role, I architected ML projects ensuring my teams were making steady progress towards meeting client expectations.
Recently, I've been building generative AI (Stable Diffusion, LLMs) and vision (CLIP, BLIP, ViT) projects. I am specifically interested in multimodal, ensamble, and augmentation research. See my GitHub for more projects!
I am seeking projects and work in these areas. If you would like to collaborate with me on a project, message me below!
Research paper on a model that predicts text prompts from generated images using an ensemble of multimodal models, including CLIP, BLIP, and ViT. Uses a high-quality dataset and of 100k custom generated images, cleaned to have low semantic similarity.
View PaperDeep learning project and paper to forecast world events outside of the training data using large language models (LLMs) including GPT-3, GPT-3.5, and GPT-4. Prompt engineering techniques including few-shot learning, expert identification, and inference averaging are employed to increase accuracy.
View PaperAn adversary emulation framework with special features including use of a messaging app as a communication channel, a hybrid encryption scheme, and shell code injection. The implant is served by a containerized listening server, MySQL database, and Flask app.
Visit ReadmeA vision transformer based machine learning pipeline built to classify plant specimens and organize records, built for the Harvard Herbarium & BU Spark!. The pipeline is composed of an ensamble of fine-tuned OCR and NER models along with a ViT plant classifier to pick correct classifications in the correct context.
Visit RepoDeep learning project to create a model that finds similar videos using a dataset of 1000s of videos. Scored 8th place of 212 participants in Meta AI’s Video Similarity Challenge in the more difficult "matching" track.
See LeaderboardTail-recursive interpreter designed to parse, push, and execute 12 unique commands on a custom made stack. Compiler implemented to compile a traditional programming language into custom-built stack language.
Visit Repo