As part of building NiBL, a restaurant-focused delivery startup, I designed a scalable, real-time food delivery platform architecture tailored for the U.S. market. I structured the system around a microservices-based backend with an API Gateway handling routing, authentication, and rate limiting, and defined core services for users, restaurants, orders, payments, driver dispatch, and real-time tracking. I implemented a geospatially-aware driver matching system using Redis to efficiently assign orders based on proximity, reliability, and load, while modeling the order lifecycle as an event-driven state machine to ensure consistency across services. To support live tracking, I designed a high-frequency GPS ingestion pipeline with optimized protocols (SSE for customers, WebSockets for drivers) and a tiered storage strategy for performance and cost efficiency. The architecture leverages PostgreSQL for transactional consistency, Redis for caching and geo queries, and Kafka for asynchronous event streaming, with containerized deployment on Kubernetes to enable autoscaling and fault tolerance.

This is the freelance videography portfolio website for a professional videographer. We worked closely to pick the right design and features to showcase his work and services. I built with Next.js and TailwindCSS while using CapCut to edit the main video on the homepage. We then finalized photos and videos for each portfolio section and hosted it on Vercel.

My colleague and I built this project because we were curious about how city complaints and neighborhood conditions actually affect real estate value in NYC. Instead of just analyzing a static dataset, we wanted to create something interactive — a system where we could store, process, and explore both 311 complaints and property sales in one place. So we designed a full-stack platform using SQL, Python, and Flask that lets us analyze trends, compare neighborhoods, and visualize how complaint volume correlates with housing prices.

As someone with interest in finance and personal banking, I wanted to explore how machine learning could help predict loan approvals based on customer data. I gathered a dataset of bank customers, cleaned and preprocessed the data using Pandas, and then built classification models with scikit-learn to predict whether a loan would be approved or denied. I evaluated different algorithms, tuned hyperparameters, and visualized the results to understand which factors most influenced loan decisions. This project helped me apply data science techniques to a real-world financial problem while deepening my understanding of ML workflows.

As part of building a data-driven restaurant discovery startup, I wanted to understand which restaurants in NYC's East Village were genuinely buzzing — not just highly rated, but actively talked about across multiple platforms. I built a multi-source data pipeline in Python that pulls restaurant data from Google Places, Yelp, and Reddit, then combines them into a single composite buzz score using weighted, normalized signals. The score factors in ratings, review volume, and organic community mentions. I also built an interactive Plotly dashboard in a Jupyter notebook to visualize the rankings, rating vs. popularity breakdowns, and price tier analysis — designed to be shared with the team and presented to potential investors as a clean, code-free HTML report.