Open to opportunities

kevin dsouza

@kevindsouza

Message

Generalist solving challenging problems at the intersection of AI and science.

Canada

Message

What I'm looking for

I seek a role that fosters innovation and collaboration, focusing on AI advancements.

I am a dedicated researcher and problem solver with a strong background in AI, science, and computation. My journey began with a B.Tech in Electronics and Communication Engineering, followed by advanced degrees from the University of British Columbia, where I earned both a Ph.D. and an M.A.Sc in Electrical and Computer Engineering. My academic pursuits have been complemented by hands-on experience in various roles, including postdoctoral fellowships and internships, where I have developed innovative solutions in machine learning and data science.

Currently, I am a Postdoctoral Fellow at the University of Waterloo, collaborating with the Royal Bank of Canada on cutting-edge projects involving large language models and evolutionary algorithms. My previous role as a Technology Consultant at a stealth startup allowed me to enhance the factual accuracy of LLMs, while my tenure as a Machine Learning Engineer at Gandeeva Therapeutics resulted in the creation of a deep learning-based protein design tool that achieved state-of-the-art success in drug discovery. I thrive in collaborative environments and take pride in leading research teams to achieve impactful results.

Experience

Work history, roles, and key accomplishments

Current

Postdoctoral Fellow

Current

University of Waterloo, Royal Bank of Canada

Oct 2023 - Present (1 year 8 months)

Working on LLM's + evolutionary algorithms for aligning decentralized agents and Reinforcement learning with verifiable rewards (RLVR) for spatial optimization using heuristics. Leading the research team and managing everyone's data and ML workflows/projects.

LLMs)Reinforcement Learning Data Management Workflows

Technology Consultant

Stealth Startup

May 2023 - Sep 2023 (4 months)

Worked on improving factual accuracy in LLMs. See thoughts compiled later here.

LLMs)Natural Language Processing Research Data Analysis

Machine Learning Engineer

Gandeeva Therapeutics

Feb 2023 - May 2023 (3 months)

Developed deep learning based protein design tool for drug discovery. The designed protein sequences achieved SOTA success rate in the lab. Gandeeva currently uses this platform for their discovery campaigns.

Deep Learning Drug Discovery Machine Learning Python PyTorch

Ph.D. Candidate

University of British Columbia

Jan 2019 - Jan 2023 (4 years)

Designed RNN based representation learning methods for biology, which achieved SOTA on multiple downstream tasks at the time.

RNN Representation Learning Biology Machine Learning Deep Learning Research

Machine Learning Intern

Gandeeva Therapeutics

Jun 2022 - Dec 2022 (6 months)

Developed DL-based pipeline for target specific antibody design.

Deep Learning Machine Learning Python PyTorch

Machine Learning Intern

Skycope Technologies

May 2018 - Sep 2018 (4 months)

Built Skycope's data and machine learning infrastructure from scratch and modified Faster-RCNN to successfully detect and locate drone signals in the spectrogram. Integrated ML software into Skycope's existing software stack. Skycope is now using this as a flagship service in their automatic drone detection software.

Data Infrastructure Machine Learning Faster RCNN Drone Detection Software Integration Python Linux

Education

Degrees, certifications, and relevant coursework

National Institute of Technology Karnataka

B.Tech, Electronics and Communication Engineering

Completed a Bachelor of Technology in Electronics and Communication Engineering. Gained foundational knowledge in electronics, communication systems, and related engineering principles.

University of British Columbia

Ph.D., Electrical and Computer Engineering

Conducted advanced research in Electrical and Computer Engineering, focusing on AI, science, and computation. Designed RNN-based representation learning methods for biology, achieving state-of-the-art results on multiple downstream tasks.