Multimodal Content Moderation
Group Course Project,
Developed a state-of-the-art multimodal (image and text) content moderation model that performed well on Meta’s Hateful Memes Challenge for the company Clarifai. Built a computation pipeline to augment and generate synthetic text and image data samples using Stable Diffusion for training neural networks.
Cloud Database System
Group Course Project,
Built a distributed database system inspired by Amazon’s Dynamo in Java with multiple servers that could be dynamically removed or added. System featured data replication across servers, heart beat for server failure detection, and gossiping to achieve eventual consistency.
A Material Point Method for Snow Simulation
Individual Course Project,
Reimplemented the paper A Material Point Method For Snow Simulation in 2D using Python and Taichi Lang.
Portfolio item number 1
Project,
Short description of portfolio item number 1
Portfolio item number 2
Project,
Short description of portfolio item number 2
Compositional Premise Retrieval
Group Course Project,
Developed a novel method for training Transformer models for retrieving premises for Automatic Theorem Proving. Benchmarked baseline retrieval methods: TF-IDF, BM25, and Sentence-BERT.
Animated Wallpaper Generation
Personal Project,
Generated animated desktop wallpapers with open-source state-of-the-art AI models. Used AI software and compute infrastructure including ComfyUI, Hugging Face, and Modal.
