TIL 2020 is a competition organized by the Defence Science and Technology Agency as part of BrainHack2020 which involves Natural Language Processing and Computer Vision to train a robot to complete challenges autonomously.

Details:
• Achieved 1st place out of 92 Uni/Poly teams.
• Implemented transfer learning to create object detection models.
• Adapted image augmentation script into data pipeline to enhance model training.
• Developed code logic and applied object detection models in robot to autonomously identify objects, navigate obstacles and retrieve target.

RoboMaster is an annual robotics competition held in Shenzhen, China and is organized by DJI. The purpose of RoboMaster is to develop young talents in the field of robotics and engineering.

Details:
• Part of the standard robot development team.
• Developed real-time software architecture for robots to interface with different sensors and motors.
• Implemented control loops for precise movement control.
• Embedded systems development in an ARM Cortex microcontroller.

Problem Statement:
How might we leverage a network of smart infrastructure in the built environment to make better and more timely sense of emergency incidents (e.g. detection of fires developing, building collapses, falls, road traffic accidents etc.) and to trigger early intervention measures, without the need to activate precious emergency resources?

Solution:
Our solution aims to notify drivers in advance and make the journey smoother and more efficient for the emergency vehicles.
• Using the streetlights as visual signals for drivers to keep to the side of the roads.
• Preemptive control of traffic lights to prepare an uninterrupted route to the scene of an emergency.

• Achieved 4th place

Orgnized by the IEEE NUS Student Branch, the Hackathon aims to provide a platform for cross-disciplinary appreciation and project implementation (hardware and software) with an end goal of exposing the varsity landscape to hack culture.

Problem Statement:
Using design concepts, create a software or hardware product that can benefit and contribute to environmental change in your community.

Problem identified - Irresponsible pet keeping results in undesirable release of invasive species into the ecosystem, endangering local wildlife.
Identified HMW statement - How might we use technology to create an edutainment tool to teach proper petkeeping skills to young children?

Solution - FLOMK [Free Landscape Of Meddling Kritters]:
• An interactive pet that helps to simulate the experience of raising a pet.
• Used arduino for robotic control.
• LED with photodetector to simulate feeding the pet.
• Object detection for responsive behaviour to changes in the environment.
• Robotic pet will be used in conjunction with a mobile app.

PRISM is a weekend-long social hackathon and festival that aims to provide a platform for youths to generate innovative, and sustainable, solutions to challenge statements sourced from community partners.

Solution - Turnout:
• A platform that provides informal avenues to connect people who need mental support to those who are willing to interact and share
• A platform to connect two categories of people.
• Partake in discounted activities together.
• Conversation prompts that encourage sharing.
• To overcome the stigma behind seeking help.

Design-Centric Programme Project

Details:
• A motorized trolley attachment that allows trolleys to be handled more easily.
• Aims to reduce the physical intensity of labour intensive jobs and to ease integration of the ageing population into the workforce.

Introduction to Design Thinking - Project 2

Details:
• To reduce waste in food packaging.
• Defining the problem through user interviews.
• Ideation with insights from users' perception of value.
• Created a live-sized prototype for product demonstration

Introduction to Design Thinking - Project 1

Details:
• To revolutionaize transport via a central platform.
• Conducted users' needs analysis.
• Applied the Double Diamond Model.
• Divergent thinking to explore different possibilities.
• Convergent thinking to focus on an issue and create a targeted solution.

A mini project to experiment with pix2pix generation using CNNs

Details:
• Canny edge detection method to extract edges from the original images to simulate simpler sketches.
• U-Net model to create an output image of same size as input. (224,224,3)
• VGG16 network as a feature extractor to compare the generated image and the original image.
• Root mean squared distance between the features as a loss.

This project explores the use of transfer learning to create a classifier for images of gym equipment and deploying it on a web page.

Motivation:
• When the thought of losing weight and getting in shape presents itself, a common go-to place that many think of is the local gym.
• However, these new gym-goers who enter the place for the first time face a very similar experience.
• Their untrained eyes breeze past a sea of foreign objects as they head towards the only equipment they know: dumbbells and treadmills.
• The occasional motivation to try out new machines might surface, only to be quickly brushed aside by the intimidating looks of massive bulking beasts or nonchalant individuals with their earpieces plugged in.

If only there was an easier way to make sense of these complicated machinery and learn how to use them...

This project explores the use of K-means clustering to identify the most common colours in an image, and replaces all pixels with the closest colour.

Motivation:
• I find the minimalist version of pictures to have a unique and pleasing aesthetic.

Final project for a course in Datascience with Python.

Motivation:
• Kickstarter is a platform where practically anyone can submit a project of their own interest, with minimal barriers to entry.
• This analysis give us insight towards the types of products and/or services that have greater market interest.

This was a short project done to explore the use of Google's Dialogflow for chatbots.

Details:
• The content surrounds a simple Q&A regarding COVID-19 and circuit breaker measures in Singapore.
• Written in Node.js and deployed on Heroku.