San Francisco, CA O-1A Work Authorization (Extraordinary Ability)
Back to home

Academic & Personal Projects

A selection of projects from USC graduate coursework in AI, ML, and NLP, and GUC undergraduate work in biomedical signal processing, game development, and blockchain.

Alzheimer's Disease (AD) Classification

2023
  • Achieved SOTA competitive results with 74.8% recall by novel approach of considering volume correlations with multi-atlas spatio-contextual GINs in 3-way AD from structural MRI images classification on ADNI dataset split.
  • Reproduced (Lee and Lee, 2021) work as part of a team of 2 to detect emotions in conversations, employing pre-trained language models without requiring structured knowledge bases.
  • Generated 20% of 2 datasets based on English-Arabic subtitles alignment and showed comparable accuracies of both to English datasets in evaluation.
  • Reproduced (Chen et al., 2014) with ARC-Standard and ARC-Eager transition-based parsing with PyTorch on CoNLL data and experimented with learning word embeddings.
  • Implemented an intelligent Go-game agent with adversarial search — minimax with alpha-beta pruning — as part of USC CSCI 561 Foundations of AI.
  • Implemented BFS, UCS, and A* search to solve 3D mazes by the best route, as part of USC CSCI 561 Foundations of AI.
  • Built a web scraper to assemble a topic-labeled URL corpus, extracted TF-IDF features, trained a Bayesian classifier, and packaged the model for deployment as a Flask API.
  • Modeled movement direction from EEG spike trains of 36 neurons using a KNN classifier.

EMG Decomposition

2020
  • Pre-processed needle EMG signals and decomposed them into constituent MUAP trains using standard decomposition algorithms.
  • Pre-processed raw ECG signals into clear QRS complexes and implemented a Sinus Arrhythmia detector.
  • Implemented a ScroogeCoin-style blockchain simulator in Python covering transaction validation and chain state.
  • Game built in C#/Unity during the Brackeys Game Jam 2020.1.
  • Ranked best project in class, modeled as Constraint Satisfaction Problem (CSP) in Prolog CLPFD, interfacing with Django API.
  • Architected as model engine and query formatter. SQL tables are created, then Prolog queries are formatted and issued.