Hello! and Welcome!

Team project for UMass COMPSCI 603 Robotics. Built an autonomous symbol-search system for the Triton mobile robot that fuses a live camera stream and LiDAR with trained CNN/MobileNetV2 classifiers to find and drive toward printed geometric symbols. Migrated the original ROS1 Noetic stack to ROS2 Jazzy on Ubuntu 24.04 while preserving the trained PyTorch checkpoints, and validated the pipeline with unit tests, lint checks, and headless end-to-end simulation in Gazebo Harmonic.

End-to-end Monte Carlo Localization for a TurtleBot3 with a 2D LiDAR, porting a COMPSCI 603 ROS1 implementation to ROS2 Jazzy. Implements a vectorized probabilistic odometry motion model and a likelihood-field sensor model built from an occupancy map, then runs a full particle-filter cycle of predict, correct, estimate, and resample. Packaged as a reproducible Ubuntu 24.04 / Gazebo Harmonic Docker container that streams Gazebo and RViz to macOS through a browser-based noVNC desktop.

Trains simulated mobile robots to follow a right-hand wall using tabular Q-learning and SARSA, reducing a noisy 360-degree LiDAR stream to a compact discrete state and learning a policy over repeated episodes. Shows the full engineering progression from an original ROS1 Stingray implementation to a reproducible, Dockerized ROS2 Jazzy and Gazebo Harmonic redesign for macOS, including 500-episode reward histories and learning-curve analysis for both algorithms.

Ultimate Map for Academic Strategy & Success (UMASS) is a comprehensive web platform designed to help university students thrive academically and prepare for their future careers. Students may share resources, access course information, evaluate professors, collaborate with peers, and strategize their academic journey.

This project trains a series of next-token language models, starting from a linear predictor through multilayer perceptrons, self-attention, and small transformers. Character-level experiments use Tiny Shakespeare, while word-level experiments adapt the same transformer backbone to WikiText-2 and Penn Treebank (PTB). Hyperparameter sweeps compare context length, embedding size, depth/width, attention heads, optimizers, schedulers, and batch sizes, with results logged to report_src/ alongside plots and generated text samples.

Coursework for CSCI 689 implementing normalizing flows (RealNVP) and diffusion models on a 2D toy dataset. Code generates PDFs/plots documenting each question’s results.

A learning progression through non-blocking, reactive backend development in Java, from raw Mono/Flux primitives in Project Reactor up to a three-service microservice system that orchestrates a purchase order across two independent databases over WebClient. Covers backpressure, hot vs. cold publishers, scheduler selection, retry/timeout policy, and bridging blocking JDBC code into a non-blocking pipeline, tested with StepVerifier, virtual-time tests, and WebTestClient slice tests.

This is an advance algorithm project from CS course - Algorithms and Data Structures.
Algorithm Deliverable D - finding a good Ham cycle path by local search or Genetic Algorithm.
Algorithm Deliverable C - From the starting city use depth first search with iterative deepening to find the nearest goal city.
Algorithm Deliverable B - find the shortest Bitonic Tour of the cities.

This Project is Full-Stack Application of Airline Search service Restful API.
Client side, single Web page, was implement with Angular CLI version 8.3.20.
Server side implements by Java spring boot and connect to MySQL Database.

An end-to-end TensorFlow/Keras comparison of eight neural network architectures on grayscale CIFAR-10, spanning single-layer and deep perceptrons, CNNs, dropout regularization, a VGG-style stack, and a residual network (ResNet). Each architecture is trained under the same data pipeline across SGD, SGD with momentum, and Adam optimizers to study how architectural choices change learning under identical conditions.

This mobile app is for practice react native, react reduxe, and google firebase that implement by android studio, etc.

Parallel and Distributed Algorithms --- ICS 440-01 Fall 2021
Creating multi-thread of self-running object to wait for client
requests to be sent up then send back the appropriate response.
Use the writeUTF(String s) method on DataOutputStream to send a reqeust
(or a response) over the OutputStream of the Socket.
Use the readUTF() method on DataInputStream to receive a request (or a
response) from the InputStream of the Socket.