- Created a custom dataset of 2,000+ labeled ASL hand signs using OpenCV and MediaPipe, capturing 100-frame batches via webcam with real-time hand tracking and automatic CSV logging for training.

- Trained a neural network on MediaPipe keypoints using TensorFlow and Scikit-learn, achieving 90%+ accuracy across test sets and real-world validation under varied lighting and pose conditions.

- Developed a real-time recognition pipeline that loads the saved model and predicts live hand signs only when the hand's motion stabilizes, reducing noisy predictions and boosting reliability.

- Integrated a user-facing buffer system: recognized signs accumulate in memory and can be spoken aloud using text-to-speech via spacebar input on a separate thread.

- Built a dynamic interface for users to adjust model confidence thresholds in real-time, enabling usability even with lower-certainty gestures or partial signing.

- Visit the website here!

- Developed an interactive educational tool in Elm to visualize 23 algorithms including sorting, searching, tree traversals, graph search, and heap operations.

- Platform is actively used by UT-Arlington faculty in CS instruction, both in live lecture demos and self-paced study material.

- Leveraged functional programming principles (immutability, pure functions, modularity, etc.) to create a maintainable and testable codebase.

- Designed the UI for clarity and accessibility, with intuitive controls to pause, reset, and inspect each stage of algorithm execution.

- Used SVG rendering and custom animations to present each algorithm step-by-step.

- Built entirely with Elm, HTML, CSS, and SVG (no JavaScript), demonstrating Elm's power in functional web development.

- Visit the website here!

- Designed and implemented a web-based puzzle game where players must place queens on a colored, region-based board so that no two queens attack one other.

- Created a bespoke board generation algorithm with a custon flood-fill technique, ensuring exactly one solution per puzzle.

- Developed an interactive SVG-based UI using JavaScript to support queen placement, note-taking, save states, difficulty leveling, and more.

- Built with FastAPI and Pydantic to serve dynamic puzzle data and validate all user inputs, leading to robust and secure API handling.

- Employed Postman to test and document API functionality; implemented rate limiting to protect endpoints from abuse and ensure consistent performance under load.

- Runs as a Dockerized full-stack web service on Koyeb, with automatic redeployment triggered on every GitHub push through CI/CD integration.

- Built a step-by-step walkthrough for implementing a feedforward neural network from scratch using Python and NumPy, without relying on high-level ML libraries or frameworks.

- Trained the network on the MNIST handwritten digits dataset, achieving 95%+ classification accuracy through custom architecture design, careful weight initialization, and manual hyperparameter tuning.

- Explained the full learning pipeline in a code-first, math-supported format that covers forward propagation, backpropagation, gradient descent, activation functions, and loss minimization in detail.

- Included Matplotlib visualizations of the training process, showing loss curves, weight evolution, and accuracy improvements across epochs to provide insight into model behavior.

- Designed the project as a beginner-friendly learning resource for students and developers interested in demystifying how neural networks work beneath the surface.

- Developed a security-focused REST API to manage a phonebook database, featuring rigorous input validation for names and international phone numbers using regular expressions.

- Built with FastAPI, SQLite, and SQLAlchemy, utilizing parameterized queries to protect against SQL injection attacks.

- Implemented JWT-based authentication and role-based authorization (read vs. read/write) with token expiration control.

- Created a persistent audit log system that timestamps and tracks every user action in rolling `.csv` files for scalability and compliance.

- Deployed in a self-contained Docker container with mounted volumes for data and logs; fully testable via an automated Postman collection.

- GitHub repository kept private to maintain the integrity of the class project.

- Built a compiler for the SPL (Simple Programming Language) using Scala, CUP (parser generator), and JFlex (lexer).

- Implemented a full pipeline: lexical analysis, recursive-descent parsing, type checking, and intermediate code generation.

- Developed robust input validation rules from a formal grammar and integrated detailed error diagnostics.

- Included a full symbol table with scope tracking and function call semantics for nested procedures and recursion.

- Generated IR code (intermediate representation) with frame pointer management, memory allocation, static links, and control flow translation.

- GitHub repository kept private to maintain the integrity of the class project.

- Managed Java WebSocket connections and JSON strings to facilitate real-time, 2-4 player word search games.

- Authored comprehensive JUnit test suties to automate verification of game logic via JUnit and Github Actions.

- Mapped out detailed project requirements using Excel and designed system architecture with UML diagrams.

- Developed responsive front-end interfaces in HTML & CSS, seamlessly integrated with a WebSocket-powered back-end.

- Utilized Maven for project management, developed the codebase using VS Code, and deployed the game on an online server.

- Built a multipurpose Discord bot tailored for ACM @ UT Arlington (1400+ member server) using OpenAI's GPT-4.1, GPT-01-mini, DALL·E 2, and the Discord API.

- Offered real-time Q&A, AI image generation, and campus-specific academic resource navigation.

- Integrated slash-commands (/resources, /ask, /draw, etc.) with persistent context support to streamline user interactions.

- Deployed via Heroku using worker dynos and a GitHub-powered CI/CD pipeline; managed all secrets through encrypted config vars.

- Added security by restricting bot responses to authorized servers only and blocking all direct-message interactions to prevent unauthorized usage.

- Achieved a 61% reduction in storage by reducing the original 10,304-dimensional image vectors to 150 principal components.

- Implemented PCA by computing the covariance matrix of the dataset, followed by eigenvalue decomposition to identify top principal components.

- Reconstructed facial images retained 98% visual similarity to the original data despite compression, as measured by the Structural Similarity Index (SSIM).

- Designed visualizations for the mean face, top eigenfaces, and reconstructed images, effectively communicating the primary steps in PCA.

- Designed a user-friendly to-do list application with a built-in calendar view for task organization and scheduling.

- Developed a FastAPI backend for managing tasks with a MongoDB database, supporting CRUD operations.

- Integrated dynamic calendar rendering with automatic month navigation and date-specific task filtering.

- Containerized the entire application using Docker for seamless deployment across multiple environments.

- Engineered custom malloc and free implementations using four different allocation algorithms.

- Integrated advanced memory management techniques, including splitting and coalescing blocks, efficiently handling variable-sized memory requests and minimize fragmentation.

- Benchmarked allocation algorithms, comparing performance against the standard system malloc to evaluate efficiency and charted results.

- Employed debugging tools like GDB to troubleshoot memory leaks and ensure stability during extensive use cases.

- Developed a file compression tool using Huffman coding, achieving efficient, lossless reduction in file sizes.

- Programmed an XOR-based encryption system with user-defined keys for securing files.

- Assembled algorithms and data structures, including min heaps, binary trees, and dynamic arrays.

- Automated file naming with suffixes for clarity and ease of use.

- Ensured robust memory management to maintain program stability and prevent memory leaks.

- Created an interactive family tree visualization with a scalable and pannable canvas, intuitively allowing users to explore large family trees.

- Integrated an SQLite database for efficient data storage, ensuring seamless retrieval and persistence of family member details and their relationships.

- Designed a comprehensive graphical user interface (GUI) using Tkinter to manage and visualize familial relationships.

- Leveraged data structures like dictionaries and lists to efficiently manage the hierarchy of family members and their interconnections.

- Deployed a custom Ubuntu shell, enabling command-line interface operations, including command execution and redirection.

- Designed a built-in command interpreter to handle essential shell commands such as cd and exit, improving usability and system control.

- Utilized low-level system calls like fork(), exec(), and wait() to create a functional, multi-process environment within the shell.

- Developed a multi-page website using HTML, CSS to showcase my professional experience and projects.

- Designed a dynamic landing page featuring highlights and easy navigation.

- Created experience and portfolio pages showcasing my work/volunteer experience and programming projects, respectively.

- Curated a gallery page displaying photographs I have personally taken, enhanced with consistent formatting regardless of photo size.

- Built an interactive educational tool to demonstrate monoalphabetic and transposition ciphersb using Java's AWT and Swing libraries.

- Implemented a dynamic, user-friendly GUI that allows users to input text and visualize how different ciphers encrypt and decrypt messages in real-time.

- Incorporated data structures such as HashMap, ArrayList, and matrices to efficiently map cipher keys, manage dynamic text data, and perform complex transposition ciphers.

- Included detailed explanations and step-by-step visuals within the interface to enhance user understanding of encryption concepts.