My Projects

Postare a full-stack web application that serves as a visual discovery engine, allowing users to explore, save, and organize image posts seamlessly. The platform ensures a secure and intuitive environment for users to share and discover creative ideas.

Tools Used:

Backend: Go, Java Spring, PostgreSQL

Frontend: Next.js, CharkraUI

Cloud: AWS RDS, ECR and EC2

Postare a Kotlin based Android mobile application that functions as music player, where users can find, listen to, and favorite their own songs and albums.

Tools Used:

Backend: Kotlin, Room Database

Frontend: Android Jetpack, Google Exoplayer

Vegrendishop is a simple e-commerce platform for handcrafts. Built as a full-stack web application, it allows users to explore, purchase, and post their own crafts easily. The platform ensures a secure environment to purchase (ensured with StripeAPI) and discover different crafts.

Tools Used:

Backend: Go, ElasticSearch

Frontend: React, AntDesign

Cloud: Google Cloud

Pocket Twitch Player is a full-stack web player that recommends personalized twitch contents to the users, which allows users to search for Twitch resources such as streams, videos, and save them for future usage. Personalized recommendations are added to improve user experience.

Tools Used:

Backend: Java SpringBoot, MySQL

Frontend: React, AntDesign

Cloud: Amazon RDS

This project uses a Spark-based machine learning algorithm to categorize users according to their YouTube video comments. The dataset is processed users comments via RegexTokenizer and Word2Vec in SparkML. The adopted Logistic Regression and Random Forest models are tuned via k-fold cross-validation. TF-IDF methodology is applied for feature extraction and implemented an unsupervised Latent Dirichlet Allocation (LDA) model to identify the top 5 topics among the target user group.

Tools Used:

Programming & Tools: Python, Spark, Numpy, Pandas, Seaborn, Matplotlib

Modeling: Tokenizers, TF-IDF and LDA

This projects uses San Francisco Crime data from 2003 to present to study and model crimes in the area, using big data OLAP techniques with Spark, to discover possible interesting trends as well as developing possible models for the number of crimes in the area. During the process, a data processing pipeline is set up using Spark RDD, a K-means model and a time series model are established to model the number of crimes in the area in terms of geographical and temporal relationships.

Tools Used:

Programming & Tools: Python, Spark, Numpy, Pandas, Seaborn, Matplotlib

Modeling: ARIMA and K- means model

This project adopts the UNet++ and transUNet model and improves the existing architecture, by leveraging an addition of Atrious Spatial Pyramid Pooling (ASPP) model at the end going through the output and integrating spatial channels attention blocks within the decoder for these two models, respectively. The model results are tested using a combined loss of Cross Entropy and Dice loss. The result shows improvement using the improved architecture.

Tools Used:

Programming: Python, PyTorch

Modeling: UNet, ASPP

This project builds a deep learning time series model (LSTM) to predict the price of stocks on TensorFlow. The time series data is segmented into different length of sequences and standardized the prices. The LSTM model parameters including activations function, learning rate, and drop-out rate are tuned via TensorFlow on GPU. The resulting model is tested on the stock of American Airlines (AAL) and reached the best performance with a MSE of 0.002

Tools Used:

Programming & Tools: Python, Spark, Numpy, Pandas, Seaborn, Matplotlib

Modeling: ARIMA and K- means model