Projects

Merged 3 nationwide datasets (33M congestion rows, 46-col accidents, 22-col pollution) into a unified state-day panel; explored how congestion relates to accidents, seasonality, and emissions using statistical models & geospatial plots.

Tool that runs multiple LLMs on the same dataset and reports confidence, calibration (ECE), and an aggregated ensemble prediction. Includes CSV uploads and a simple React + Node/Python workflow.

Full-stack app that tests how inflation data (CPI) relates to short-term S&P 500 returns. Includes a FastAPI backend for models and a Next.js dashboard for running scenarios.

Designed a normalized SQL Server schema and analytic queries to study how student social media use relates to sleep, mental health, relationships, and academics.

Implemented two representations of the Deque abstract data type: an array-based version with front/back indices and a node-based version using sentinel nodes. Focused on correctness, invariants, and performance.

Built multiple implementations of the autocomplete operation and compared how different data structures handle prefix queries, sorting, and returning the top matches.

Implemented multiple versions of a priority queue and compared how their different representations affect operations like remove-smallest and changing priorities.

Implemented shortest-path algorithms on weighted directed graphs and applied the same ideas to find minimum-energy seams for image resizing.

Used pandas and linear regression to explore housing data and build models that predict house prices.

Used product review data from Amazon.com, turned reviews into word-count features, and trained logistic regression models to predict whether a review is positive or negative.

Used LendingClub loan data to predict whether a loan is safe (+1) or risky (-1) with decision trees and a simple random forest, and compared training vs validation accuracy across different tree depths.

Built and trained several PyTorch neural networks (NetA–NetD) on CIFAR-10 to classify 32×32 color images into 10 classes, comparing fully connected and convolutional models using GPU training.

Implemented k-means in NumPy and applied it to TF-IDF vectors of ~5.9k Wikipedia biographies to study clustering behavior under different inits and K values.

Modeled ~119k April 30, 2020 COVID-era tweets with TF-IDF + NMF to discover latent topics, inspect top words per topic, and analyze tweet–topic weights and outliers.

RMarkdown analysis tying S&P 500 trends to CPI, tax burden, and housing costs (US vs non-US), with fully reproducible data cleaning, merging, and visualization.

MC3 over add/remove-one neighborhoods; rcdd linearity checks; MH with neighbor-count correction; compared to greedy selections.

Posterior mode via Newton–Raphson; Laplace approximation; MH sampler; parallelized 60 fits with snow; posterior means + MLE sanity checks.

Two high-perf C/C++ versions (LAPACKE, GSL) for LM marginal likelihood; GEMM/solve/log-det with careful memory/layout; matched R baseline and spec.

13-process simulation (referee + 12 players) with point-to-point messages; rally probabilities; compact payloads; scoring/set logic; clean termination.