Khalil Virji

Machine Learning Researcher - University of California, Irvine (July 2024 - Dec 2025)

Research on advancing deep learning models for atmospheric forecasting, aiming to extend the skilful prediction window beyond 14 days.
Extended state-of-the-art models to better capture slow-moving atmospheric processes, improving long-range forecast stability.
Evaluated global numerical forecasting models against deep learning models for their ability to capture the Madden-Julian Oscillation (MJO), finding clear improvements in learning-based models despite not being explicitly trained for MJO prediction.
Studied state-of-the-art atmospheric foundation models based on vision transformers, graph neural networks, and diffusion.
Enabled distributed training of 1.2B+ parameter models on limited hardware using memory-efficient data streaming and lightweight fine-tuning techniques (e.g., LoRA).
Supervised by Dr. Isabella Velicogna in the Department of Earth System Science at UCI.
Tools used: Python, PyTorch Lightning, NumPy, Xarray

Robotics Researcher - McGill Mobile Robotics Lab (September 2022 - June 2024)

Graduate research under the supervision of Dr. Gregory Dudek.
Published research on robust perception and control methods for autonomous robots operating in real-world environments.
Built a vision-based robotic system for autonomous scuba diver following and recovery. The system combined YOLOv7 for diver detection, Simple Online and Realtime Tracking (SORT) for tracking, and spiral search for diver recovery. Investigated both deep reinforcement learning and traditional PID control paradigms and validated the system through pool and open ocean deployments.
Trained a transformer network to predict future terrain conditions from visual input and candidate action sequences, enabling an autonomous vehicle to remain on favorable terrain while navigating mapless, unstructured environments.
Transferred deep reinforcement learning control policies from simulation to physical robots, leveraging domain randomization and adversarial learning to ensure robust sim-to-real transfer.
Tools used: Python, C++, PyTorch, ROS2, Unity, Docker

Lead Software Developer - Compost Montreal (June 2022 - June 2024)

Lead developer responsible for maintaining and enhancing a Shopify e-commerce store.
Managed end-to-end storefront development, from feature implementation to site maintenance and troubleshooting.
Eliminated over 30% of recurring SaaS costs by replacing third-party services with in-house solutions.
Restructured bilingual marketing and optimized SEO localization for English- and French-speaking customers.
Helped the team with compost collection runs to gain a holistic perspective on the company's mission.
Additional tools used: JavaScript, HTML, CSS, Liquid

Software Developer Intern - ColorKarma (May 2022 - September 2022)

Worked as the primary developer on a web application aimed at fostering connections between creative designers and manufacturers.
Managed an overseas front-end development team, improving my ability to clearly articulate technical ideas.
Improved the readability and scalability of the API by establishing a structured code paradigm, incorporating a route-controller-service architecture.
Tools used: Node.js, Express.js, MySQL, Angular

Teaching Assistant - McGill University (January 2022 - May 2022)

Software Developer Intern - CloudOps (May 2021 - September 2021)

Got exposure to full stack development by creating backend API features using Java and Spring Boot, and UI features using Vue.js.
Worked with testing frameworks such as Jest, Spock, and JUnit.
Became proficient in writing queries to a MySQL database and Elasticsearch.
Participated in design discussions and helped create mockups, models, and documentation.
Gained experience with professional development practices including PR reviews, code walkthroughs, acceptance reviews, and tasking sessions.
Abstracted AWS, Azure, and GCP services into a single platform to streamline multi-cloud data management.
Launched a self-service API trial feature for prospective clients with scoped permissions and expiration controls.
Additional tools used: Docker, JIRA

Web Developer Intern - Sensequake (May 2020 - September 2020)

Created a web portal for clients using Node.js, Express.js, MongoDB, HTML, and CSS. This reduced data visualization latency by over 50% and enabled more efficient interpretation of raw sensor data.
Enabled real-time structural health monitoring by building web commands to communicate with low-level vibration sensors.
Worked with AWS services including EC2, S3, and Lambda.
Implemented an automatic modal analysis algorithm from scratch using techniques from selected research papers.
Additional tools used: Plotly, Three.js, MATLAB, Python

Software Developer Intern - Traction on Demand (January 2020 - March 2020)

Developed interactive Lightning Web Components, automated workflows, and analysis algorithms for clients.
Studied the Salesforce platform in depth with an emphasis on application development, customization, security, and database architectures.
Became proficient in writing Apex code and SOQL queries.
Participated in code walkthroughs, PR reviews, and tasking sessions.
Earned over 50 Trailhead badges and became Salesforce Platform Developer 1 certified.

Teaching Assistant - McGill University (September 2019 - January 2020)

Computer Science Tutor - Computer Science Undergraduate Society (January 2019 - January 2020)

Tutored students for COMP 202, COMP 206, COMP 250, COMP 273, COMP 251.
Taught core CS concepts such as object-oriented programming, algorithms, data structures, and systems programming in Java, C, and Python.
Held weekly office hours to assist students with coursework.

Tutor - Nimbus Learning (January 2018 - September 2019)

Organized one-on-one tutoring sessions with McGill, Concordia, and CEGEP students.
Helped students with core courses in Mathematics, Computer Science, and Chemistry.
Tutored over 80 hours with a 4.95/5 tutor rating.

Supervisor - Pedalheads (May 2016 - September 2018)