Projects

Academic/Professional Projects

1. Machine Learning-Based Malware Detection and Attack Challenge

Multi-stage challenge on ML-based malware detection and adversarial evasion attacks.

  • Built ML models for malware detection with strict constraints (≤1GB RAM, ≤5s response).
  • Developed evasive malware binaries to bypass detection, testing robustness of defense models.
  • Explored both black-box and white-box attack/defense strategies.
  • URL: GitHub Repository

2. Improving CLIP Training (CSCE636 Deep Learning Project)

Deep Learning project on Contrastive Language-Image Pretraining (CLIP).

  • Implemented and compared 5 loss functions (CLIP, SogCLR, CyCLIP, VICReg, OnlineCLR).
  • Evaluated with 3 optimizers (AdamW, SGD, Adam).
  • Trained on CC3M subset (100k); validated on MSCOCO & ImageNet.
  • Results showed CyCLIP with AdamW achieved best average performance (16.9%).
  • Dataset: CC3M, MSCOCO, ImageNet

3. Deep Learning Coursework Projects

Regression and Optimization

  • Ridge Regression & Lasso on E2006-tfidf dataset, with cross-validation and error curve analysis.
  • Implemented SGD for Logistic Regression with step decay, cosine decay, and polynomial decay learning rate schedules.

Logistic Regression & MNIST Classification

  • Proved convexity of regularized logistic regression objective.
  • Derived Lagrange dual for distributionally robust optimization.
  • Built multi-class logistic regression model for MNIST with minibatch SGD, tested weight decay, and analyzed overfitting.

4. Natural Language Processing Projects (CSCE638 NLP, Spring 2025)

  • Tokenization & Byte-Pair Encoding: Built subword vocabulary from text corpus.
  • Positional Encoding: Constructed and visualized sinusoidal positional encodings.
  • Advanced Text Classification: Built CNN and fine-tuned BERT on Yelp Reviews dataset with GloVe embeddings.
  • Seq2Seq with LSTM: Implemented encoder-decoder with attention for Quora Question Pairs, with GloVe initialization and text generation.

5. Software Security Projects

Coursework projects using C, C++, and Python.

  • Buffer Overflow & ROP Attacks: Exploited vulnerable C programs to hijack control flow.
  • Format-String Attacks: Crafted malicious printf payloads for information leaks.
  • Memory Protection (Bound Checking, CFI, SFI): Implemented runtime defenses.
  • Fuzz Testing & Symbolic Execution: Used fuzzers and symbolic execution tools to find vulnerabilities.
  • Malware Analysis: Reverse-engineered and analyzed malicious binaries.

6. Configuring Hadoop Cluster and Executing Big Data Applications with Apache Spark

  • Configured a Hadoop cluster using HDFS and Spark.
  • Implemented small-scale big data applications for distributed storage and parallel computation.

7. User Folder Migration

  • Built a migration system for 200+ enterprise clients, transitioning thousands of user reports, graphs, and imports into a new structure.
  • Used Blue-Green deployment strategy to ensure zero downtime.

8. Enhancing Vending Machine Application with Microservices and Containerization

  • Refactored vending machine app into microservices.
  • Used Docker for scalable deployment and management.

9. AI-Powered Reversi Game

  • Designed an AI-driven Reversi game with Java Swing GUI.
  • Implemented Alpha-Beta Pruning algorithm for strategic computer moves.

10. 4-bit Computer System Design

  • Designed a SAP-based 4-bit computer system.
  • Implemented 28 assembly instructions and simulated in Proteus Design Suite.

11. Hall Management System

  • Built a hall office management system with:
    • Desktop app: Java Swing GUI.
    • Web app: Laravel + MySQL backend.
  • Won 2nd Runner-up at NSysS 2016 project show.

12. 29 Card Game

  • Multiplayer 29 card game built using Java multithreading + socket programming.
  • Implemented server-client model for real-time gameplay.