Projects

GPT Code Clippy

  • Built GPT Code Clippy which is an open-source version of GitHub’s AI pair programmer, Copilot. We have managed to create a dataset for code generation from scratch, produce detailed documentation on our model development and evaluation and finally a VSCode extension for users to try our AI pair programmer

  • As of Oct 1 2021, we have over 300+ stars on GitHub

PromptSource

  • Major contributor to PromptSource, a toolkit for collecting and applying templates of prompting instances

  • Part of my open source language model efforts in BigScience on Multitask Prompted Training

Metastatic.ai

  • Metastatic.ai is a metastatic cancer early detection platform

  • Was developed as an AI startup project focusing on using Computer Vision to detect metastatic cancer early with support of the Department of Computer Science Innovation Lab (DCSIL), University of Toronto

  • A fully working prototype can be found at https://narwhals-ai.herokuapp.com/

Melanoma Detection

  • Built a UNet-based ensemble model for performing skin lesion segmentation for Melanoma Detection with 94.1% accuracy

  • Emerged winner of the Medical Imaging Challenge, MIT Critical Data Hackathon 2019


Deepore

  • Studied a deep learning model by the researchers in University of Queensland known as the Chiron model which is used to reduce the errors that occur during the base calling procedure performed by the Oxford MINion DNA sequencing device. We attempted to improve upon the Chiron model and create our own model, Deepore which provides more accurate results

  • Work presented at the National University of Singapore's 11th School of Computing Term Project Showcase (STePS)