I’m an aerospace engineering PhD student at Georgia Tech. I’m also a MS student in CS. My research involves optimization of space situational awareness architectures. I am also broadly interested in deep learning, statistical modeling, and reinforcement learning.
Projects
Here are some of the projects I’ve enjoyed working on. Click on them to learn more!
SoundStream: An End-to-End Neural Audio Codec in PyTorch
Implemented and trained SoundStream from scratch, a neural audio codec forming the foundation for audio tokenizers used in SOTA audio modeling. We provide modules for causal convolutions, vector quantization, and all discriminators in addition to the SoundStream model itself. Training is fast, taking approximately 8 hours on a single NVIDIA L40S GPU. Click on the image to learn more!
ModernBERT for the Patent Classification Task
ModernBERT for complex patent classification, demonstrating >2x faster inference than traditional BERT with state-of-the-art precision and F1 scores on a domain test set. Introduced USPTO-3M, a large public dataset of 3 million patents.
On Device LLM Inference
Built a library from scratch in C++ for constructing the GPT2 language model. KV Caching and cache friendly access patterns accelerate tokens/sec. Inference is done locally on an M1 chip! Half precision GELU operations yield a 1.27x inference speedup.
On Device Character Recognition
Built a library from scratch in C++ for constructing convolutional neural networks. Trained on MNIST dataset. Again, real time inference on an M1 chip.