AvatarsFTW

October 2024 to Present

3D Human Avatars From the Wild

Developed pipeline to improve performance of SOTA 3D Human Avatar Generation pipelines on in-the-wild images.

»

▸ Created an inpainting pipeline consisting of segmentation, keypoint estimation, diffusion, and GAN inversion to remove occlusions and generate missing human-body areas.

▸ Developed novel method for human keypoint estimation to condition diffusion models for human image generation.

▸ Integrated PyMAF-X and incorporated additional losses to improve SMPL-X regression and improve body fit for highly dynamic poses.

Overview diagram of AvatarsFTW. Results showing improved normal maps due to the inpainting pipeline. Results showing improved SMPL-X fit due to the new regression and iterative fitting process.

Neural Plane Optimization

April 2024 to May 2024

Faster UAV Design with Machine Learning

Used neural networks and non-linear optimization to determine unmanned aerial vehicle design parameters.

»

▸ Utilized a neural network and particle swarm optimization (PSO) to speed up my AeroNUS 2022 aeroplane optimization program over 100 times.

▸ Increased design space by over 1000 times by training a DNN with Keras for 2D to 3D aerodynamic estimation.

Optimized airfoil design for the given flight mission requirements. Particle swarm optimization mean and best values, showing quick convergence in a large design space.

vidi

June 2023

for UC Berkeley AI Hackathon 2023

Designed a smart, AI-powered glasses to improve accessibility for individuals with visual impairment in a team of 2.

»

▸ Designed glasses in a team of 2 to aid blind individuals with a Raspberry Pi, speaker, microphone, and camera.

▸ Wrote software to simultaneously process sensor inputs and outputs using YOLOv8, GPT-4, and multiprocessing.

First time assembling all the electronics together. Struggling to fit all the wires in such a small box! Final product with everything assembled.

EG3301R

January 2022 to November 2022

Reforestation Robot

Worked in a team of 3 to design and build a land vehicle to autonomously replant tree seeds in post-wildfire forests for an iDP module, EG3301R.

»

▸ Worked with CAL FIRE as a customer to gain problem insights and design specifications.

▸ Used Solidworks to produce CAD model for chassis, optimizing parts for ease of manufacturing.

▸ Used ArduRover and an arduino to design controls for the robot.

▸ Designed and soldered power harness.

▸ Prototyping using 3D printing and laser cutting.

First prototype of power harness along with controls. This is a bench test to ensure the motors function as intended. The first arm blade manufactured and mounted, with sufficient electronics for testing.

AeroNUS

July 2021 to April 2022

Chief Engineer

I was the chief engineer for the AeroNUS UAV Competition Team, where we designed and built an Unmanned Aerial Vehicle (UAV) for the American Institute of Aeronautics and Astronautics (AIAA) Design, Build, Fly (DBF) Competition in 2022. We managed to achieve the highest report ranking in NUS history, at 14th place out of 110 teams. In this position, I oversaw the various the subteams and coordinated joint design efforts, and led manufacturing, scheduling, and team organization.

»

▸ Used XFLR5 and AVL to analyze airfoils, wing and empennage configurations, and determine key stability values and trim conditions.

▸ Designed wings, tail, and fuselage using Solidworks, managed full CAD model for final plane.

▸ Performed Solidworks static and topology FEA simulations using nonuniform force distributions to minimize the weight of our wings.

▸ Wrote a Multi-Disciplinary Design Optimization (MDO) program using MATLAB to determine plane sizing for our prototype and final planes.

▸ Developed python scripts to write G-Code for precise foam CNC manufacturing of lifting bodies.

▸ Manufactured planes using foam hot wire CNC, carbon fiber prepreg and molding, CNC milling, laser cutting, and 3D printing, along with hand manufacturing with various glues, foams, and wood.

▸ Designed and manufactured the controls circuit and wrote and implemented an Arduino program to sequentially deploy payloads.

▸ Organized the team, planned the overall schedule and key deadlines, coordinated meetings, and helped resolve disputes.

Using a laser cut jig to create bent stringers for our prototype fuselage. We did not use this process in our final plane because the manufacturing time was not worth the gain in aerodynamics. Here is our fuselage in three parts, which was necessary to take our plane in flight-legal boxes to the US. We attempted creating Carbon-Fiber landing gear. Along with that, some carbon fiber spars can be seen in the background. Here is a picture of our final team in the US.

PPE Detection with Invigilo AI

August 2021 to December 2021

Machine Learning Research Project

Implemented various CV models using Pytorch and YOLOv5 for helmet detection in construction sites under the guidance of startup Invigilo AI. Counted as UROPS (EG2605).

»

Languages: Python

Packages: Pytorch, OpenCV

▸ Worked with staff from Invigilo AI to find students interested in the project.

▸ Used OpenCV to test the effectiveness of various image pre-processing techniques for CNNs.

Kaliber Labs

January 2023 to July 2024

Software Engineering Intern | AI Researcher

At Kaliber Labs, I spent one and half years as a part of the AI research team, working with image generation and classification, large language models, and 2D and 3D computer vision.

»

Languages: Python, C, C++, Bash

Packages: Tensorflow, Keras, PyTorch, OpenCV, Nerfstudio, React, NextJS, Flask, AWS

▸ Fine-tuned Llama-3-8B on surgical data, achieving 84% of GPT-4o’s performance for 2% of the operational cost.

▸ Accelerated convergence of camera pose estimation algorithm by 10 times with particle swarm optimization (PSO).

▸ Developed intra-operative camera registration algorithm taking under 10 seconds utilizing neural radiance fields.

▸ Implemented methods for digital linear and geodesic measurements between points with above 90% accuracy.

▸ Spearheaded development and prompt engineering for a new multi-modal patient chat-bot- combining surgical image model outputs, RAG with patient data, and LLMs- with Flask, NextJS, AWS, and Langchain.

▸ Fine-tuned stable diffusion to generate synthetic training data, expanding surgical tool datasets by up to 5 times.

▸ Employed synthetic data generation and augmentation to train multiple image classifiers to above 97% accuracy.

▸ Trained pix2pix for modality conversion and added interpolation pipeline to multiply MRI dataset size by 10 times.

F-Drones

May 2022 to September 2022

Unmanned Aerial Systems Engineering Intern

At F-Drones, I primarily focused on communication systems and testing and development of autopilot software.

»

▸ Created Lua scripts and modified Ardupilot code for additional battery and EKF functionality.

▸ Setup SITL (Simulation-in-the-Loop), defined key operational SOP, and used it for testing geofences, autopilot bugs, failsafes, and scripts.

▸ Used 3DS Max and Solidworks to create SITL models of our aircraft.

▸ Researched and tested analog, digital, and 4G communications and setup 4G modules currently used for deliveries. Collaborated with partner companies and led colleagues to perform tests.

▸ Created mapping tool with Google Maps API to track network strength for various 4G providers across operational areas in Singapore.

SciFie Robotics

March 2021 to July 2021

Robotics Intern

Conducted R&D, prototyping, and design for a magnetically-mounted robot for surface preparation, paint application, and NDT of large metal structures with a team of 6.

»

▸ Developed engineering sizing spreadsheets based on specifications of robot modules.

▸ Designed and modelled a magnetic mounting bracket, NDT mounting and couplant delivery assembly, and hydroblasting arm capable of sustaining 10’000 PSI water pressure using Solidworks.

▸ Created engineering drawings in Solidworks and performed GD&T.

▸ Collaborated with various manufacturers and suppliers to create and procure components.

▸ Improved a WAGO PLC control system and interface for movement and controls of the robot arm.

▸ Wrote first draft of a business plan and designed a website with Wix.

Carnegie Mellon University

2024-2025

MS in Computer Vision


Perception Research at CMU AirLab

National University of Singapore

2020-2024

B. Eng (Hons.) in Mechanical Engineering

Highest Distinction

2nd Major in Innovation and Design


Chief Engineer and Mentor, AeroNUS

Vice President, Innovation and Design Programme Students' Club

Projects Director, Aviation Club

iDP Students' Club

January 2021 to December 2022

Vice President External, formerly Head of Professional Development

Spent two years in the iDP Students' Club as a way to involve myself better with the community of my second major.

»

▸ Influenced club direction to focus on improving network with alumni and startup founders.

▸ Assisted subcommittees with logistics, external liaising, and planning for various events, such as the IDEATE 2022 ideathon, UTR Design Fest, and Summer Internship Program.

▸ Directly involved in planning and logistics of iDP Carnival, attracting over 100 students, and a brand new iDP Freshman Orientation Programme.

▸ Previously led a team of 5 in the Professional Development Committee to run an internship program for iDP students, having sourced over 60 internships by working with over 20 companies and various NUS parties.

▸ Organized an industry talk with Apple and contacted judges and sponsors for the IDEATE 2021 ideathon and spearheaded logistics for the judging procedure with 11 judges and 150 students.

AeroNUS

May 2022 to December 2022

Mentor

After our tenure in AeroNUS, we organised the first AeroNUS bootcamp and created a structured training phase, enabling the new team to achieve a record world #1 ranking for their proposal, and improve on our overall competition result by 15 spots.

»

▸ Overhauled advertising and recruitment process, bringing in over 55 signups, improving on last year's 30 signups.

▸ Planned and hosted the AeroNUS bootcamp, a hybrid, month-long introductory course for new students interested in joining the team for over 40 participants.

▸ Mentored the new team of 20 in creating their first plane.

▸ Updated MDO sizing program for 2023 AIAA DBF rules and implemented lifting line theory for more precise CL/CD calculations.

Performing outreach at Science Center to build community interest in aviation. Analyzing plane crash and teaching new team about stress concentrations and building for failure. Our first workshop!

Software

PyTorch, Tensorflow, and Keras

Prompt Engineering

LLM Fine-Tuning

Image Generation and Classification

Linux

Python

Javascript

Flask, React & NextJS

AWS

Git

Hardware & Robotics

Solidworks | CSWP Certified

Mechanical Design, Sizing and Optimization

Manufacturing and Machining

Aerodynamics, Sizing, and Design

Circuit Design

ROS

Raspberry Pi

Arduino