Aidan Dempster
Teaching Assistant at University of Toronto
Toronto, Ontario, Canada
Summary
🤩
Rockstar🎓
Top SchoolAidan Dempster is a robotics engineering undergraduate at the University of Toronto with a decade of hands-on software experience spanning web development, computer vision, and embedded robotics. He has delivered production front-end features at EvoText, built landmark-detection data pipelines and GUI tools for medical research at UHN, and extended flight-control and simulation tooling at Cleo Robotics. Currently a TA and undergraduate researcher in bionics, he studies EEG decoding for motor intention prediction while contributing to large-scale open-source AI work. Notably, he helped scale a DALL·E 2 reproduction for LAION by writing a distributed training script for a 5B-parameter model with Hugging Face Accelerate and sorting the 2B-image LAION2B dataset using PySpark on a 96-core cluster. Comfortable across Python, PyTorch, React, and cloud/distributed systems, he bridges research-grade ML with production engineering. Colleagues describe him as the kind of engineer who moves smoothly between GUI polish, data engineering, and low-level control systems.
10 years of coding experience
2 years of employment as a software developer
Bachelor of Applied Science - BASc, Robotics Engineering, Bachelor of Applied Science - BASc, Robotics Engineering at University of Toronto
Stackoverflow
41reputation
3kreached
1answer
3questions
Github Skills (18)
pytorch10
artificial-intelligence10
data-pipelines10
machine-learning10
webdataset10
deep-learning10
data-pipeline10
text-to-image9
python8
babeljs6
curl6
php6
mp46
ffmpegjs6
transport-stream6
Programming languages (7)
Github contributions (5)
lucidrains/DALLE2-pytorch
May 2022 - Aug 2022
Implementation of DALL-E 2, OpenAI's updated text-to-image synthesis neural network, in Pytorch
Role in this project:
ML Engineer
ML EngineerContributions:20 reviews, 20 commits, 22 PRs in 3 months
Contributions summary:Aidan primarily focused on developing and improving the data loading and training pipelines for the DALL-E 2 implementation. Their contributions included creating a custom dataloader for training the decoder, incorporating embedding insertion from external sources, and integrating webdataset for efficient data handling. The user also implemented a flexible framework that can train the decoder with different sources of image embeddings, including clip generated embeddings and precomputed embeddings.
pytorchdall-eimage-synthesisdeep-learningsynthesis
Veldrovive/video-processor
Nov 2019 - Feb 2021
Contributions:55 commits, 44 pushes, 2 branches in 1 year 3 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.