• I've been working as an Intern here since 1st Nov 2021
• I've done ad-hoc tasks such as making a program to web-scrape multiple websites and importing the information into an excel sheet
• I've also done tutorial videos, help centre articles and tech support, helping users out with whatever problem they may have
• Through this experience, I learnt how truly important accountability and communication is
• My position at the shop includes being a cashier, cook and a barista
• Though my time there was short, I learnt many important values such as teamwork and communication
• I understood the value of communication and teamwork as the cashier, barista and the cook has to communicate with each other if not there would not be sufficient resources to serve customers and the service would stop
• Developed, tested and analysed two methods of detecting blurred images
• Cleaned and modified datasets for testing and training purposes (e.g. gaussian blurring images, reformatting images)
• Researched different edge operators for development of feature engineering paired Support Vector Machine model for classification
• Tested various configurations/developments of Convolutional Neural Networks for optimized performance in this problem (e.g MobileNetV2)
• Made detailed reports on developments in research every week and presented report to supervisor (worked with numbers and graphs)
• I was a speaker in the event
• I spoke and presented my project, SwordTale, a RPG(roleplaying game) integrated into a Discord bot which has been worked on for about 6 months with the help of a lecturer, Mr Gi Soong Chee and two JC students, Rui Yang and Isaac
• I spoke about important aspects of the code like the combat mechanism which used asyncio
• I was a participant in Shopee Code League 2021 and 2022. Shopee Code League is a 2 week event where many varying workshops are held for participants to learn. 3 competitions are held in the 2 weeks to test what the participants have learnt. The event was open to students and progessionals and allowed for teams.
Shopee Code League 2021
• Through the various competitions, I had learnt many new concepts such as Natural Language Processing, Sentiment Analysis, Data Processing, Algorithms and basic Machine Learning
• The first competition was on Data Anaytics "Multi-Channel Contacts Problem Statement" which gave us a dataset with columns ticketid, email, phone, order id, contacts of customers. We were then required to find tickets that are related either directly or indirectly through Email, Phone or Order ID and submit a csv file with two columns, one for all the ticket ids and another for the ticket-trace which is a string of all related ticket numbers separated by dashes and the total amount of related tickets at the back, separated by a comma.
• The second competition was on Data Science "Address Elements Extraction" where we were given a dataset of addresses to build a model to correctly extract Point of Interest (POI) Names and Street Names from unformatted Indonesia addresses.
• The third competition was just a competitive programming competition
Shopee Code League 2022
• The format of SCL was that of codeforces contests but they still held workshops for us to learn more about competitive programming
• Through the workshops we learnt more about topics such as dynamic programming, recursion, graph algorithms and many more
• I managed to help my team solve one and a half questions
• This was an 8 day long competition where we had to use AISG's building blocks, PeekingDuck/TagUI/SG-NLP to design and build a tool that can help, assist, inform or entertain Singaporeans or businesses
• I made a convolutional neural network for the competition which takes in pictures or videos as an input and draws bounding boxes around the humans in the picture and labels the boxes as "fake" if it predicts that certain human had been deepfaked or "real" otherwise
• I had spent majority of the time on pre-processing the images from Google's FaceForensics++ dataset
• I first went through all the pictures in the dataset and used a pretrained model to detect human faces in the picture, since there were too many data for my personal laptop to handle, I deleted the pictures with no human faces or really poorly detected human faces
• To further reduce the amount of computation needed, I made a function to go through and crop the pictures in a way that the faces were perfectly in the middle and they were the specified size (480 by 480)
• I then made a basic convolutional neural network with 5 convolutional layers using tensorflow with a maximum accuracy of 58%
• I attended this 6 hour workshop held by Google to learn Google Cloud fundamentals.
• We got exposure and learnt how to use many technologies including Kubernetes, Firebase and more