Selected work
Things I've built.
A mix of coursework, hackathon projects, and independent work, spanning full-stack systems, applied ML, and real-time interactive apps.
01
Capstone · E. & J. Gallo Winery
Gallo Sanitation Management System
Backend & data engineering
AWS Bedrock
Full-stack capstone
A full-stack engineering capstone built with E. & J. Gallo Winery to replace a paper-based sanitation tracking system used in production facilities. The original workflow relied on manual checklists and spreadsheet entry, which created inefficiencies and limited real-time visibility. Our system digitizes this into a centralized web application that standardizes sanitation workflows, improves traceability, and supports audit and compliance needs.
I worked primarily as a backend developer and data engineer, focusing on system architecture, database design, and core application logic. That included building the submission and inspection systems, designing the facility-to-production-line data hierarchy, and integrating frontend and backend APIs. I also implemented an AI-powered admin assistant using AWS Bedrock.
02
Independent research · Interpretability
Visualizing Emergent Concepts in Transformer Hidden States
DistilGPT2
PCA · t-SNE · UMAP
3D visualization
I analyzed how DistilGPT2 represents and organizes information internally by extracting hidden state activations across transformer layers. The goal was to better understand how concepts form and evolve inside the model, since these representations are high-dimensional and not directly interpretable.
To make these representations visible, I applied PCA, t-SNE, and UMAP to project them into 3D space, allowing me to study how different concepts are structured and how interpretability changes depending on the visualization method.
03
Independent project · Computer architecture
RISC-V CPU Simulator
Python
RISC-V ISA
Datapath design
A Python simulation of RISC-V CPU hardware, implemented in both single-cycle and multi-cycle versions, that decodes and executes compiled machine code.
I implemented the fetch-decode-execute-memory-writeback datapath, including instruction decoding, a 32-register file, data hazard detection that inserts stall cycles to resolve read-after-write dependencies, and pipeline flushing to handle control hazards from taken branches and jumps.
04
Hacktually Hackathon · UC Merced
Merced EMS Staging Optimizer
Flask
React
Applied ML / spatial optimization
A web application designed to help emergency medical services determine optimal ambulance staging locations across Merced, California. Users input the number of available units and an optimization goal, such as minimizing response time based on current call demand. The system visualizes key locations on an interactive map with recommended staging points and explanations for each placement.
Built with a Flask backend and a React-based heatmap interface, modeling emergency call density using neighborhood-level data with time-based weighting to generate staging recommendations that balance coverage while avoiding redundant positioning. Developed during the Hacktually hackathon, exploring how AI can support spatial decision-making.
05
Team project · 3 engineers
Competitive Drawing Application
Real-time multiplayer
Canvas & input handling
A multiplayer drawing game where players join lobbies and compete in real-time drawing challenges, sketching from prompts and submitting results within each round. Built in a team of three.
I focused on the core drawing engine: a freeform canvas, pen tool, eraser, undo system, color selection, and adjustable brush sizes, the main interaction layer players use every round.