About Me
Hi! I am a recent Stanford graduate (masters 2023, bachelors 2022) in computer science with a focus in AI and security. I'm interested in software architecture, AI systems, and data-intensive problems. In my free time, I like to surf, ski and fly planes.
Recent Projects / Papers
Visualizing Attention in Vision TransformersPaperPoster
For my CS231N Final Project, we built a vision transformer (ViT) from scratch and visualized attention maps across a series of hyperparameters.
Interpretating Prompt-Tuning Across SuperGLUE TasksPaper
For my CS224U final project, we explored the interpretability of finetuned prompts. We the looked at prompt stochasticity under different conditions and finally investigated the zero-shot task transferability of learned prompts.
Vector Similarity Search: A Deeper DivePaper
For my CS168 final project, I wrote a research survey on the frontiers of sub-O(n) vector similarity search.
ChatAWG (Asymetric Warfare Group). Hacking for Defense Project
This was a small project I created for my team during Hacking for Defense under Steve Blank and Joe Felter. This is a simple webapp that accepts uploaded PDFs via a Steamlit frontend and then uses Llama-Index to parse and query the document via an LLM. This aids military members in training by making textbooks searchable.
Read it, or Reddit? A GNN Approach to Predicting Relationships on RedditGoogle Colab
This was the final report for CS224W, Machine Learning with Graphs. Brandon Kang and I built a simple Reddit Post reccomendation engine using Graphical Neural Networks in Pytorch.
An End-To-End Pipeline for Pose Estimation in Mobile Applications Using Kalman FilteringPaper
For my final project of CS231A we investigated using three different models for pose estimation. Each model was used in an end-to-end pipeline application that started on a phone, went to a cloud computer running a Flask server to run the model, which then sent the results back to the device. We additionally investigated using denoising and a Kalman Filter and report the results.
Leave It To BERT: Exploring Methods for Robust Multi-Task PerformancePaperPoster
For my CS224N Final Project, we built smaller version of BERT, imported pretrained weights for BERT, then experimented with finetuning different superstructures on top of BERT to acheive good generalization on three different tasks.
CNN Plays Geoguessr: Transfer Learning on ResNet50 for Classifying Street View ImagesPaperPoster
For my CS229 final project, we finetuned ResNet-50 to play Geoguessr, a game where you guess the country given a Google Street View image, and acheived better than human performance.
Encrypted Chat Client
Encrypted chat client based on Signal's Double-Ratchet Algorithm.
Decentralized Exchange (DEX)
A simple DEX modeled off of Uniswap V2.
Password Manager
Secure password manager based on the JS subtle.crypto library.