Project Portfolio
A collection of personal and collaborative projects I have worked on over the years
A full-stack platform that scores an investment portfolio's risk and explains the reasoning in plain English, built for new and conservative investors. Deployed and multi-user, with per-symbol signals, a portfolio-level risk grade, and a what-if trade simulator.
Key Features
- Composite Buy/Hold/Sell signal from 9 technical indicators (RSI, MACD, Bollinger, ATR, beta, Sharpe, Sortino, and more) with a per-indicator contribution breakdown
- Portfolio risk grade (A–F) from concentration, correlation, effective leverage, beta, and historical stress tests
- What-if trade simulator showing before/after risk without touching the real portfolio
- Plain-English risk explainer with an LLM layer guarded by a validator and deterministic fallback
- Multi-tenant auth, self-hosted on Oracle Cloud via Docker with GitHub Actions CI/CD
Spotify Dashboard — Music Analytics
A self-hosted alternative to Spotify Wrapped: a deployed dashboard that surfaces top artists, tracks, and listening stats across multiple time ranges on any day of the year rather than once every November. Next.js frontend over a Flask backend that owns all Spotify OAuth and API access, shipped as two Docker images to an Oracle Cloud VM.
Note on the live site: Spotify caps apps in Developer Mode at five authorized listeners, and lifting that cap requires a quota-extension approval that is not granted to personal projects. The deployment is fully public and the login flow works, but only allowlisted Spotify accounts can complete authorization — so expect the sign-in step to reject you unless I have added your account. The source and the blog series are the best way to see the whole thing.
Key Features
- Full Spotify OAuth flow handled server-side, with tokens never exposed to the browser and same-origin API proxying through Next.js rewrites
- Top artists, tracks, playlists, and listening profile across 4-week, 6-month, and all-time windows
- Album explorer panel that opens any top track’s album and badges the tracks that also chart in your top 50, with ID-first matching and a disclosed title-match fallback
- Normalizer layer that absorbs Spotify’s API deprecations in one place, keeping routes thin after endpoint removals and response fields disappearing upstream
- Self-hosted on an Oracle Cloud VM behind a shared nginx edge proxy and Cloudflare, with GitHub Actions building and publishing both images on every push
- CI runs pytest against captured real API response shapes plus TypeScript typechecking and a production build, so the next upstream schema change fails the build instead of silently emptying a chart
- In progress: a vibe-based recommendation engine using crowd tags and LLM inference to hold mood axes (tempo, energy, language, familiarity) independent, replacing the retired Spotify audio-features and recommendations endpoints
Predictive Modeling of Aortic Root Dilation
Undergraduate research at the Cardiovascular Imaging Research Lab (CVIRL) on predictive modeling of aortic root dilation in pediatric patients with cardiovascular diseases. Selected for the Purdue Summer Undergraduate Research Fellowship (SURF) program, Summer 2026.
Key Contributions
- Construct and annotate echocardiogram datasets for automated aortic wall segmentation and diameter measurement
- Develop machine learning models to predict progression of aortic root dilation and classify disease subtypes
- Integrate data annotation and modeling pipelines to improve scalability and reproducibility
Mead Johnson PeptideX
A Python-based peptide database and Next.js web application developed in collaboration with Mead Johnson through The Data Mine. Enables biomedical researchers to search and query peptides with relevant data visualizations.
Key Features
- Integrated data from multiple open-source peptide databases, consolidating structural, biochemical, and trait-specific characteristics
- React web interface for researchers to search and query peptides
- Database infrastructure with SQLite for improved portability and deployment
SASE Sniping Discord Bot
A Discord bot that sends automated real-time notifications about nearby SASE mentors based on their class schedule. Helps members stay informed and maximizes efficiency in gaining points in a friendly competition.
Key Features
- Real-time monitoring and instant Discord alert notifications
- Improves efficiency with an automated process instead of buried messages
BucketList Community App
A $1,000 commissioned goal-focused community platform developed for Purdue ACM SIGAPP. Features community boards, goal tracking, and social interaction capabilities to help users achieve their personal and professional objectives.
Commercial Success
- Delivered commissioned mobile application for client business
- Developed Community Board feature with React Native
- Implemented scalable architecture for community engagement
Led development of an alumni networking platform connecting current students with graduates for mentorship and career guidance. Managed a student development team and coordinated with 40+ community members.
Leadership & Impact
- Led student development team through full project lifecycle
- Coordinated requirements gathering with 40+ stakeholders
- Deployed production-ready application serving the school community
- Implemented secure authentication and user management
An ML-powered mental health journaling application built with TypeScript and React. Features sentiment analysis to analyze writing trends and track mental health symptoms over time, helping users gain insights into their emotional patterns.
Key Features
- Real-time sentiment analysis of journal entries
- Trend visualization and mental health tracking
- Secure data storage with Firebase
- Responsive design for cross-platform accessibility
A comprehensive stock-monitoring application featuring real-time data tracking and machine learning-powered price predictions. Implemented multiple ML models to provide investment insights and market analysis.
Technical Achievements
- Experimented with LSTM, Random Forest, and CNN models for stock price prediction
- Built scalable Flask API with PostgreSQL database
- Real-time stock data integration and monitoring
- Interactive data visualization and analysis tools
Project Categories
Machine Learning
Sentiment analysis, stock prediction models, and AI applications for disaster response and image classification.
Full-Stack Web Apps
End-to-end web applications with modern frameworks, databases, and cloud deployment.
Mobile Development
Cross-platform mobile applications with React Native and cloud backend integration.
Research & Data
Biomedical databases, data pipelines, and research applications with scientific impact.