Aqusense, Toronto, ON

May 2025 -- Present

Optimized real-time data processing pipeline by implementing TensorRT execution providers and resolving critical buffering issues, achieving consistent 40KB per epoch processing and 6-7x inference speedup from CPU baseline.
Architected cross-platform migration solution from Ubuntu 22.04 to 24.04, developing automated dependency scripts for 3+ Rust codebases and documenting 5000+ lines of code, achieving zero-downtime deployment.
Built end-to-end data collection and validation pipeline, collecting 20+ hours of structured sensor data with distance-based analysis covering multiple tracking scenarios.

University of Toronto, Toronto, ON

Jun. 2024 -- Sept. 2024

Improved learning outcomes for 200+ students in ECE344 (Operating Systems) by developing a basic shell program in C, enabling hands-on interaction with the Linux kernel and enhancing course engagement.
Increased code maintainability and modularity by implementing 3 shell features from the MIT xv6 operating system, adhering to LLVM coding standards for OS-level software development.
Enhanced shell compatibility with POSIX standards by leveraging GNU C library functions for process control and signal handling, achieving 100% compliance and ensuring memory safety.

Borealis AI, Toronto, ON

Jan. 2025 -- May 2025

Developed an autonomous RL agent for microgrid management that outperformed the baseline control system by 34% in non-renewable energy cost, while ensuring 100% load demand satisfaction during peak consumption periods.
Implemented a multi-modal time-series forecasting pipeline that integrated weather, market, and consumption data, achieving R² of 0.95 on future load predictions and improving on the current state of the art research RL algorithms.
Designed and deployed an interactive monitoring dashboard, simulating 105K+ data points from 5-minute interval recordings, resulting in 12x faster update frequency and enabling stakeholders to visualize predictive insights.

SmartMate, London, UK

Jun. 2024 -- Sept. 2024

Designed and implemented a Computer Vision model using OpenCV and PyTorch to convert ASL finger spelling into English captions, achieving a 95% validation accuracy on public ASL video datasets.
Conducted rigorous research on Continuous Sign Language Recognition (CSLR), implementing diverse architectures from 5+ research papers, including 3DCNNs, Bi-directional LSTMs, and Vision Transformers.
Authored comprehensive documentation outlining the end-to-end implementation of the ASL translation pipeline and model architecture, supporting startup pitching and funding efforts.

ML4U, Amsterdam, NL

Jun. 2024 -- Sept. 2024

Engineered a sentiment analysis model using a Decoder-Only Transformer architecture with PyTorch, spaCy, and NLTK, achieving an 87% testing accuracy on the IMDB reviews dataset.
Developed an API using the Django REST framework and PostgreSQL for database queries, enabling concurrent storage of 10,000+ customer sentiments.
Built a React-based user authentication system, shifting from server-side HTML templates to client-side rendering, reducing server response times by about 20%.

Centennial Electrical Group, Vaughan, ON

Jul. 2023 -- Aug. 2023

Fabricated and assembled HVAC control panels, including wire cutting, stripping, and termination, as well as panel punching for component mounting, using electrical engineering schematics and industry standards, resulting in the successful deployment of 50+ custom wired panels across a Toronto condo project.
Coordinated the transfer of critical electrical components like FMC conduits and copper wiring, using inventory tracking systems and logistical planning, resulting in a 15% reduction in on-site supply shortages and improved project timelines.