ideas that got a little out of hand
my digital wall of creations
Senior project · Healthcare
A mobile app for tracking IBD symptoms, medication, and patient history. My team and I built a senior-project prototype to bring daily check-ins and health history into one place. This is my favorite project I’ve built.
Bioinformatics · Full stack
I took a cancer gene research project and built a pipeline and dashboard around it. It uses gene networks to find candidates worth exploring, then lets you dig into the results.
Student consulting · Mobile full stack
Built map, search, and parking interfaces for a React Native commute-planning prototype. My team and I collaborated with Tesla stakeholders through Cal Poly’s Systems Optimization Club to bring commuting resources into one experience.
Through Cal Poly’s Systems Optimization Club, my team and I collaborated with Tesla stakeholders on a centralized commuting application for Bay Area employees. The project brought together a student development team, designers, and industry stakeholders.
The problem was fragmented transportation information and pressure on employee parking. The intended experience gave employees one place to compare driving, biking, public transit, and company shuttle options, with parking information available alongside those choices.
As a React Native frontend developer and student consultant, my responsibility was to turn the designs into usable mobile interfaces and build shared foundations that the development team could extend. Over six months, that work covered:
Colors, spacing, and typography were repeated across screens. I researched styling approaches and chose a lightweight theme-object and Context structure that fit the existing app. Components could read from the same values, so a visual change had a shared place to start.
The work involved gradually replacing duplicated styling while preserving the existing interactions. That mattered in a team project: the frontend needed to stay usable as other developers continued building.
The Figma designs repeated patterns across transportation modes, destinations, and parking. I turned those patterns into components with props for their configuration and actions, giving the team common building blocks for new screens.
The search and favorites work covered frontend interactions and sorting. Backend persistence for favorites was a separate integration concern. The map controls also began as temporary UI while the associated state logic was still being developed.
I refined the parking interface and percentage calculations so the values and interactions matched the intended design more closely. This used simple arithmetic and prototype data. The work also involved checking navigation targets and testing the interface in the iOS simulator.
The appearance-preference work established infrastructure; adoption across every screen and the final merge state were not independently verified in the supplied report.
My work focused on the React Native frontend. The broader team architecture connected that client to Node.js and Express services, Google Maps and TripShot data, and an MSSQL/Redis data layer. Those backend and infrastructure responsibilities belonged to the wider project team.
My team and I developed a centralized commuting platform and prepared it for handoff. The club’s public announcement describes routes, schedules, commute alternatives, favorites, and administrative capabilities, with future integration into Tesla One as the continuation path.
My contribution gave the frontend a shared theme, reusable components, and map, search, and parking interfaces that the team could build on. That made visual consistency and future changes easier to manage through shared code.
The platform was intended to improve access to commute resources and support alternatives to driving. A production rollout, employee adoption, and measured reductions in parking congestion or commute times were not established in the supplied evidence.
This project made me think about the people who would build on my code. A theme or component needed to make sense to the next developer as well as the screen I was working on.
Working with designers, other developers, and Tesla stakeholders taught me how much coordination goes into bringing a mobile application together. I’m proud of the interfaces I built, and of the shared foundation that made the frontend easier for the team to continue developing.
Nonprofit collaboration · Full stack web app
Built announcement creation, inbox retrieval, and recipient selection for a nonprofit’s heritage-tree inventory application. My team and I developed the platform through Hack4Impact Cal Poly.
The Central Coast Heritage Tree Foundation works to preserve and restore heritage oaks around Templeton. Its volunteer inventory work gives the project a concrete purpose: recording observations about trees that the community wants to care for.
Through Hack4Impact Cal Poly, my team and I extended an existing application with tree-entry interfaces, inventory views, geographic exploration, and volunteer announcements. The documented project team included 14 students across development, design, and project leadership.
My main engineering story was connecting announcements all the way through: selecting an audience, saving a message, retrieving it, and opening its details in the application.
As a full stack developer, my responsibility covered early tree interfaces and several iterations of the announcement workflow:
The complete tree-storage pipeline, map, authentication, and later persistent read/deletion behavior were wider team contributions. The recipient UI used an existing user-directory endpoint.
I replaced manually typed recipients with a selectable list from the application’s user directory. Administrators could choose individuals, select all users, or select administrators. The group shortcut normalized role values so capitalization did not change who was included.
The composer collected a subject and message, then submitted the selected email addresses as recipients. Those addresses identified who the in-app announcement was intended for; this was an application inbox rather than an email-delivery integration.
Directory checkboxes and group shortcuts
Sender, recipients, subject, and message
A stored announcement document
Recipient view and message details
I connected message retrieval to the inbox and replaced hardcoded content with the stored records. Opening a message displayed that announcement’s details, and the interface calculated its count from the retrieved data.
The inbox filtered messages in the browser using the signed-in person’s email. That describes the implemented display behavior; server-side recipient authorization is a separate concern.
The first message table used demonstration data for a presentation. Later changes added announcement persistence, retrieval, and directory-based recipients. Each iteration connected another part of the feature instead of treating a finished-looking table as the finished system.
That progression is what I value most about this work: taking a frontend interaction and carrying it through the API and database, then returning the saved information to the person using the app.
The three recipient options covered broadcasts, administrator updates, and selected volunteers without introducing a full chat platform or custom-group system. The directory control made the available recipients visible and removed the need to type their addresses into the form.
The Next.js application kept React pages and API routes in the same project. I connected the interface to those routes and used the announcement schema to carry sender, recipients, subject, and content into MongoDB.
An announcement naturally groups related information into one document, including recipient and read-status arrays. MongoDB and Mongoose fit that structure while allowing the models to evolve with the application’s requirements.
The broader platform also used Clerk for identity, S3 for files, and Leaflet with OpenStreetMap for geographic views. Those services explain the application’s setting; the announcement integration and earlier tree interfaces were my contribution.
I delivered merged announcement creation, stored-message retrieval, and selectable recipients, alongside the earlier tree-entry and inventory interfaces. The announcement feature progressed from demonstration content to a database-backed workflow.
For administrators, that meant choosing the intended audience from the directory and submitting a stored update. For volunteers, it meant an inbox view for announcements addressed to them. For the development team, the merged work provided a foundation for later message-state and interface improvements.
My team and I contributed to a platform that brings tree information and volunteer communication together. The foundation now links to the hosted inventory application from its public website. The public sign-in page was reachable in October 2026; authenticated use and adoption figures were not verified here.
This project gave me experience following a feature through the full stack. A form needed somewhere to save its data, and the inbox needed to show the saved message back to the right person. Getting those pieces connected was the part I’m proudest of.
Building through issues and pull-request reviews also taught me to keep the next iteration in mind. My team and I were developing for a nonprofit, and the work needed to be understandable enough for other developers to review and continue. Seeing the software connected to local conservation work made that responsibility meaningful to me.
My first Roblox game, built with my team in 24 hours at LA Hacks 2026. In Don’t Get Scammed, players set out to rescue Grandma through an obstacle course full of fake links, suspicious Robux offers, password puzzles, impersonators, and AI-generated art. Every wrong turn comes with feedback explaining what made the choice unsafe, helping kids build online-safety instincts as they play. A crash course in Roblox Studio, Lua, and just how much debugging one door can need.
PolyPrompt · Hackathon
My team and I built Task Logger at the PolyPrompt hackathon to help students divide up group projects. An LLM breaks requirements into tasks, asks clarifying questions, and helps assign work based on skills. We used AI coding agents throughout development, too—a hands-on lesson in agent orchestration, context engineering, and reviewing AI-generated code.
I worked in a team of six, Team CSA, at PolyPrompt, Cal Poly’s weekend-long OpenAI hackathon. The challenge was to create an AI-powered product for student life. In the Studying category, my team tackled the “group project nightmare”: uneven workloads, missed deadlines, and unclear responsibilities.
Dividing a group project can leave workloads uneven, dependencies unclear, and progress hard to follow. The challenge was turning a large brief into manageable tasks with clear ownership.
Task Logger uses LLM task decomposition to turn requirements into editable tasks. Skills, deadlines, and dependencies guide planning; people review the draft before locking it and running assignment.
I built project setup, skill profiles, and document-intake interfaces, then split planning into separate clarification and task-review pages. I connected the screens to existing APIs and added visible upload progress and error states.
I worked with a Codex multi-agent workflow with five main agents:
The three implementation agents worked in parallel with role-specific instructions. I also reviewed user workflows by hand and used Codex to adjust behavior; the team stepped in for manual PR review when needed.
I delivered setup, skills, upload, and task-review interfaces for the prototype. The app connects requirements to reviewed tasks, assignments, and shared workflow views, giving the project a foundation for further testing.
A photobooth that lives in your browser. Take photos, make a strip, and share it with a QR code. A little project that became one of my favorites.
Luma Leaf is a passion project that brings together two things I enjoy: making memories and building software. Inspired by Korean-style photobooths, I wanted a free browser experience that felt simple, playful, and personal—a way to turn a hangout into something you can keep.
Take four pictures with a countdown and filter, or upload four photos you already have. Then make the strip your own: rearrange the pictures, adjust their colors, choose a frame, and add a sticker theme. Download it as a PNG or create a QR link to share it with a friend.
I worked on the software: the camera flow, countdowns, filters, photostrip editor, downloads, and QR sharing. I also improved camera framing, mobile layouts, and dark mode as the project grew. Haixin Huang created the graphics that give Luma Leaf its personality.
Next.js, React, and TypeScript power the pages and editing controls; CSS Modules handle the styling. The browser’s camera API captures the photos, and Canvas combines the pictures, decorations, and frame into one downloadable image.
That keeps normal editing and PNG downloads on your device. QR sharing uses a Next.js server endpoint to temporarily hold the finished strip. Those links depend on server memory, so a restart can make them unavailable before their ten-minute expiry.
The project brings capture, editing, and export into one photo experience, with a temporary sharing option. It gives people a way to make a personalized keepsake without needing an account.
I learned more about React state, interface design, and image rendering while carrying a personal idea toward deployment. Camera framing, customization, and sharing went through several iterations. The details mattered: keeping photos consistent, carrying edits into the exported strip, and handling sharing limits.
Personal project · Productivity Pomodoro timer
A Pomodoro timer that keeps your tasks and focus sessions in one place. I built FocusFruit to make it easier to settle into a task, take a break, and come back to your progress. Your sessions stay saved, so a refresh doesn’t mean starting over.
A small pixel world built with Java and Processing, with hand-made art and characters that change as you interact.
A mobile workout planner for logging sets, reps, weights, and notes. Open it and start tracking.
Using Hidden Markov Models to explore parasite life-cycle stages through gene expression data.
Looking at how yeast changes its metabolism through gene expression, clustering, and a lot of Python.
Teaching a robot to find its way around a grid and clean it, while comparing different search algorithms.
Simulating thousands of poker hands to estimate pre-flop odds. A small excuse to play with search, probability, and Python.