Projects
Currently seeking more research opportunities to further my exploration of data in a variety of different fields.
Below are some of the research/exploration projects I have conducted:
Independent Project
FoodDetector is a one stop shop for food transparency. The main feature of this application is the use of deep learning, in particular a convolutional neural network, to detect what food a person is currently eating based upon a photo of their dish. Essentially, the application takes in a user inputted image, feeds it into the convolutional neural network, and outputs the predicted class of the image. Besides this, the application relays the key nutrient breakdown of the predicted food class to the user, and allows them to learn more about what they are eating. In this application, the user is also able to customize their age, weight, height, activity, and gender in order to get customized calories-remaining recommendations based upon the Harris–Benedict BMR formula.
Machine Learning Engineer
In order to ensure information about our on-going pandemic is inclusive and accessible to everyone regardless of ability or disability, we have created a web application that uses speech recognition, narration, deep learning, and various other tools to present the latest information about COVID-19 to users. In the last year, there have been a variety of applications developed to spread verified COVID-related information to users. Despite this, we realized a gap in catering for those with visual impairments. We decided to address this issue with the tremendous capabilities of speech recognition technology to essentially create a virtual assistant for COVID-19 information that can perform a variety of tasks ranging from answering questions about the virus using an AI-powered chatbot, to returning the top-5 global COVID-19 headlines for the current day, to even enabling quick-and-easy dissemination of COVID-19 county-specific updates to family and friends without even having to touch your phone. All of our information is continuously updated. We wanted to empower people with visual disabilities to ultimately make more informed decisions about COVID-19 for their own health and safety.
Machine Learning Engineer
This application is called HomeFinder, and is a City Search Tool that leverages the tremendous capabilities of Machine Learning. In this day and age, remote work is slowly becoming the norm. As a result, our lives are largely dictated in our places of residence and much of our time is spent in our respective communities and cities. With our algorithm, users are able to input their desired living preferences, and the web application will output the best 4 cities tailored to their needs. By maximizing the likelihood that you can call a new city you're home, we help set users up for success.
​
My role in this project was primarily to create, tune, test and finally deploy the machine learning model. After countless hours of experimentation, I decided to apply unsupervised machine learning by using K-Means Clustering to grouped cities from our dataset together based upon similarities in key features. I used metrics such as the Davies Bouldin (DBI) score to tune and optimize the hyperparameters in the model. With this trained machine learning model, we were able to input a user's desired living conditions, and output a cluster that best suits their needs. We then would deliver these fascinating insights from our ML model to the user by returning a snapshot of the top cities generated for them.
You can try it out here: https://homefinderapp.herokuapp.com
Project Lead
In the Game of Ecology problem, we were tasked with analyzing the characteristics and eco- logical interactions of three different dragons. We developed a first-order multivariable discrete time model that tracked four state variables over time: the weight of the dragon W in kilograms, its daily calorie requirement C, its height H, as well as the population P of the prey consumed by the dragon. After the construction of this baseline model, we altered parameters to represent the states of three dragons which each lived in a different climate. This allowed us to explore the impact of environmental factors on the growth of dragons. Interestingly, we found that dragons in cold, arctic environments tend to weigh more than those in warmer, arid regions. Furthermore, in larger habitats dragons can reach a greater weight due to the presence of an increased prey population. We then extended our original model to include stochasticity by modelling weight and height as random normal variables, and also considered environmental catastrophes that follow a Bernoulli distribution which affect the model parameters in different ways. Ultimately, however, the inclusion of stochasticity did not change the inherent behavior of the model.
​
Independent Project
This web application uses a tuned XGBoost Regressor trained on over 16,000 video game sales data to predict the total sales of a video game based 8 key features. The model has been deployed for public use using Heroku. With this invaluable insight, producers will be able to alter their planning and production based upon expected global sales.
​
Data Analyst
Task Magic is an application created by interns at HKMCI. This application adds gamification to task management using Asana's open API. Data analytics was carried out using Python in order to devise and develop a new methodology that redefines internal management: Internal Analytics.
As the upcoming season of the IPL draws closer, this project studies the data from previous seasons to derive fascinating insights. This is carried out through the application of unsupervised machine learning. Following this, exploration is carried out on the different clusters to derive meaningful and actionable tactical insights.
Independent Project
Independent Project
Ask a home buyer to describe their dream house, and they probably won't begin with the height of the basement ceiling or the proximity to an east-west railroad. But this Kaggle competition's dataset proves that much more influences price negotiations than the number of bedrooms or a white-picket fence.
With 79 explanatory variables describing (almost) every aspect of residential homes in Ames, Iowa, this competition challenged me to predict the final price of each home.
Independent Project
Burglaries are one of the most significant types of crimes that occur in Hong Kong every day. Mathematics may allow us to predict the frequency of burglaries each year through the analysis of past statistics and data. This allows us to make unbiased predictions of the future, by feeding numbers into complex models. The aim of this exploration is to investigate the different models for the frequency of burglaries in Hong Kong. This exploration will mainly focus on modelling burglaries in order to see if they follow a general pattern.
