Built an end-to-end battery degradation and reliability system using 185K+ NASA lithium-ion battery measurements. Developed Palantir Foundry pipelines to transform raw telemetry into battery-health and discharge features, then trained and evaluated a PyTorch degradation-risk model using leakage-safe splits, baseline comparison, and false-negative analysis. Exported the model to ONNX Runtime for local CPU inference, benchmarked latency, throughput, memory use, and model size, and added production-oriented safeguards including schema validation, inference-health telemetry, confidence-based review, failure fallbacks, and model/version traceability.
Built a full-stack child-safety platform that analyzes Roblox accounts and experiences for potentially concerning behavior and presents the findings as explainable, parent-facing risk assessments. Combined Roblox APIs, social-graph analysis, game metadata, web-scraped community evidence, deterministic risk scoring, and narrowly scoped AI-assisted evidence analysis to identify suspicious account and game signals. Developed the application with Next.js and TypeScript, authentication and data storage integrations, parent dashboards, and downloadable risk reports. The project received the Human Experience award at UWB Hacks 2026.
SupaBase
Gemini
Auth0
Web Scraping
Next.js
AWS Lyra — DubHacks 2025
WebApp Dev
ML/AI
2025/oct/
Built an LLM-powered workflow platform that helps product managers and engineering teams transform natural-language requirements into structured Agile artifacts. Developed backend, API, and model integrations using Node.js and Flask, orchestrating AWS Bedrock/Nova, Gemini, and the Strands SDK to generate consistent project-planning outputs from loosely structured business requirements. Tested the end-to-end generation workflow and integrated supporting AI services including text-to-speech capabilities.
Node.js
Python
Strands SDK
Gemini Flash API
HTML
CSS
Amazon Nova
AWS Bedrock
ElevenLabs
Flask
Gemini
Microsoft Presidio (Open Source Contributions)
Open Source
Security & Privacy
Contributed new Philippine PII-detection capabilities to Microsoft Presidio, an open-source data privacy framework. Developed and unit-tested three Python recognizers for Philippine PhilHealth IDs, passports, and license plates, implementing regex patterns, contextual detection, validation logic, and country-specific registry integration. Expanded tests, documentation, configuration, and supported-entity definitions while debugging false positives, recognizer loading, validation edge cases, and CI/test failures across the contribution workflow.
Python
Git/GitHub
Pytest
EnnyText
ML/AI
WebApp Dev
2025/present/
Originally built to help me learn Mandarin to communicate with my friend's parents, EnnyText has grown into a project that will help both high school students learn world language and help US immigrants learn English in a more intuitive and linguistically centered way. Using everything I've learned from CFA (and more through self-directed learning), my web app uses Gemini Flash 2.0 to break down Mandarin text into learner-friendly translations, with future plans for integration into Cleveland High School's Chinese curriculum and the King County Library System's ESL programs. This project is currently in progress, with many core features already shipped.
In this project, I displayed my robotics team's work through a mobile-first website, compiling our most notable projects. I composed over 50 pages of technical documents and diagrams about our team's projects, focusing on the projects developed by the computer vision sub-team I lead. I broke down our advanced computer vision, advanced physics and mathematics, and machine learning projects at the business and technical level, presented this information to judges along with the website, and won the team the Innovation and Control Award through my presentations.
HTML
CSS
JavaScript
Auto Image Labeler and Classifier WebApp UI
Robotics & CV
ML/AI
WebApp Dev
2025/jan/
I managed and oversaw the development of our image classification improvement system, owning the human-model interface design portion. This system continuously refined our image detection and classification model, which was trained on FRC 2025 robots and game pieces, streamlining the manual image labeling process for the team. I owned the human-model interface portion of the design. Our model, trained on over 170,000 images, had each one vetted for quality by the Auto Image Labeler.
Python
RoboFlow
YOLOv8
Figma
Interlink Data Transportation System
Robotics & CV
2024/mar/
I developed the Interlink System to support video duplication from multiple cameras and their transportation to different machines on board an FRC competition robot using C and GStreamer. This system allowed our robot to process the same video feed differently on machines optimized to process them; this included object detection on some machines and mapping/match logging on others.
This website helped students quickly search for Mandarin Chinese characters and stroke orders through typing them. Writing in Mandarin Chinese is notoriously difficult to master because of the vast amount of unique characters; when studying for exams, my classmates and I often had to search through textbooks, packets, and Google Searches to find out or remember how to write them. This project allowed us to streamline our studying, needing only to type the character into the site and see a visual brushstroke breakdown. This project was endorsed by my teacher and used by many Chinese students across all course levels at the institution.
HTML
CSS
JavaScript
Seattle Homelessness Project
WebApp Dev
2024/dec/
I developed this site to provide critical information about food, shelter, education, and more to the unhoused, organizing the information based on the feasibility of accessing each resource. I focused on factors such as location, site offerings, and operating hours - factors that make or break options for those with limited access to travel.
I developed this game modeling a Brazilian cafe. I made a custom cafe floorplan, along with custom sprites and assets via pixel art in MediBang. I used Unity as the game engine and C# to handle its logic, and hosted the demo online.
This project was a submission for the University of Washington WINFO design hackathon. My team and I designed and presented an app made with the purpose of helping the community through promoting healthy recycling habits. Alongside our app design and presentation, we also conducted an intercollegiate study and composed technical documentation. I owned the app design and technical documentation portions. This project won us the best presentation award and an honorable mention placement.
