/projects
A time-series econometrics pipeline examining how Google Trends-style public attention relates to Bitcoin volatility, volume, and returns.
Shows behavioral-finance thinking with VAR models, Granger causality, impulse response functions, subsample analysis, and reproducible data preprocessing.
A capstone computer engineering project: a web-based AI system that uses eye-tracking signals from a standard webcam to support autism screening.
Positioned as one of the first examples of a browser-based ASD support system using gaze capture in the browser, integrating AI prediction with full patient and test workflows.
A practical browser-based validation system for measuring the accuracy of webcam-based eye-tracking before gaze features are used downstream.
Turns browser gaze capture into a measurable signal by making calibration quality and target-cell accuracy visible instead of implicitly trusted.
A data mining pipeline that combines association rule mining and DBSCAN clustering to analyze scanpath-style eye-tracking behavior.
Shows how sequential and spatial eye-tracking data can be turned into interpretable patterns such as CTA-price co-attention and dense visual attention regions.
A reproducible machine-learning project that explains how Gini impurity drives threshold selection and decision-tree classification.
Makes decision trees more interpretable by showing threshold selection, repeated evaluation, and diagnostic performance analysis step by step.
Implemented a simulation study comparing BH, Storey's q-value, and Independent Hypothesis Weighting for large-scale multiple testing scenarios.
Showed how covariate-aware FDR methods can increase discovery power while keeping false discoveries controlled in high-throughput biological analysis.
A generic R utility for batch-converting EyeLink EDF recordings into ASC files while preserving subject/session folder structure.
Turns repetitive eye-tracking export work into a reproducible preprocessing step for downstream R workflows such as eyelinker-based analysis.
A collection of MATLAB image processing projects covering spatial enhancement, image restoration, frequency-domain filtering, segmentation, and morphology.
Implemented core computer vision operations from the pixel level up: histogram thresholding, median and min/max filters, Sobel edges, FFT-based notch filtering, Otsu segmentation, and connected components.
Built Bayesian regression and hierarchical models with brms to explain country-level happiness using GDP, social support, health, and regional structure.
Compared pooled, varying-intercept, varying-slope, and spline-based Bayesian models with posterior predictive checks, trace diagnostics, Bayesian R2, RMSE, and MAE.
Implemented a BayesFlow pipeline for a simplified SWIFT model to infer gaze-control parameters from simulated eye-tracking fixation sequences.
Turned an intractable-likelihood reading model into an amortized Bayesian inference workflow with custom simulation, adapters, neural summaries, posterior sampling, and diagnostics.
Applied PrInDT-based decision tree optimization methods to eye-tracking features and compared them against GLM and standard conditional inference trees.
Improved balanced accuracy from ~0.76 with GLM and ~0.78 with ctree to ~0.86 with optimized PrInDT-style tree models while preserving interpretability.
Built a reproducible PyTorch Lightning pipeline for six-class FashionMNIST classification, comparing CNN/MLP models and augmentation strategies.
Reached ~97.3% baseline test accuracy and showed how unseen rotations can drop performance to ~35.7%, while rotation-aware training recovers accuracy to ~95%.
Built a Java Swing desktop application for managing cows, sheep, treatments, medication records, veterinarians, farm workers, salary logic, and MySQL persistence.
Shows classic software-engineering fundamentals: object-oriented domain modeling, GUI workflows, database CRUD operations, serialization, checksum validation, and background threading.