Built and deployed an AI-powered auto video editing tool for beginner/leisure diving enthusiasts,
using state-of-the-art computer vision object detection models!
Predicted and identified the drivers of Singapore HDB resale prices (2015-2019) with 0.96 R2 & $20,000 MAE. Deployed model to web app for user price prediction.
Designed an automated workflow that covers entire ML lifecycle with CI/CD, including data validation, processing, model training, evaluation, validation and deployment.
Built a productivity tool for students and workers to easily upload their lectures/meetings and have them auto converted into summarized notes in Markdown format, which can be copied into any note-taking app.
Clustered >120k days of Fitbit data from 1.8k+ individuals to identify 4 subgroups of Singaporean working adults differentially susceptible to unhealthy behaviors.
Building a recommender system for coffee enthusiasts accompanied with a dashboard!
Trained an NLP model for generating national day song lyrics.
Building a recommendation system for Google Play games using content-based and collaborative filtering methods.
Detected Pneumonia in chest X-Ray scans using Convolutional Neural Networks and transfer learning with a F1-score of 92%.
Classify real disaster tweets using different sequence models like LSTM, Bi-directional LSTM with attention, and transformers (BERT) with best accuracy of 84%.
Detected Melanoma from skin lesion images using ensemble of EffNets and meta-data with AUROC score of 0.93.
Predicted heart disease using different machine learning models, with best model achieving 85% F1 score.
Created dashboards for visualizing datasets using Tableau.
Performed neural style transfer using VGG19 to create new pictures painted in another style.
Classified images of 120 dog breeds using Convolutional Neural Networks and transfer learning achieving 88% accuracy.
Conducted customer sales segmentation and affinity analysis on chip sales to identify groups to target for advertisements and promotions.
Simple Neural Network with one hidden layer for classifying MNIST digits with accuracy of 96%.
Trying out ARIMA forecasting of COVID-19 cases combining Google Trends search data.
Trying out forecasting of COVID-19 cases using SEIR-D model, an extension of the popular basic epidemiology SIR model.