Rayaq Siddiqui

Waterloo, ON

Algorithms, Data Structures, Operating Systems, Concurrency, Compilers, Databases, Adv. C++, Computer Vision

Visit Google →

Sunnyvale, CA

C++, Rust, TypeScript, VCS, Concurrency, Microservices, Software Design & Architecture

• Working on JJ – a modern version control system to accelerate developer productivity for Google worldwide
• Scaling distributed systems and enhancing APIs for JJ's version control operations in C++. Using concurrent operations, load balancers, various caching mechanisms, and rate limiting to deal with scale of requests
• Contributing to JJ's open source project and internal CLI in Rust. Writing comprehensive tests to ensure reliability
• Developing design documents and plans for new features to interact with various Google internal tooling
• Utilizing agent orchestration and prompt engineering to streamline software development life cycle

Visit Google →

Toronto, ON

Go, GCP, Concurrency, Distributed Computing, Databases, API, Microservices

• Working on Remote Build Execution – accelerating remote builds for clients like Chrome, Android & TensorFlow
• Utilizing globally distributed Google Cloud Platform (GCP) projects, secure internal cloud infrastructure, sharded Spanner databases, and working with multiple internal API services, following a microservice architecture
• Developing an instance monitoring system that ensures instances' migrations match the system's state

Visit d-Matrix →

Toronto, ON

C++, PyTorch, LLVM, MLIR, Convolution

• Accelerated Generative AI (LLM, SD) model inferencing by 20x focusing on compiler-level architecture in C++
• Implemented 2D Convolution, Average & Max Pooling, ResNet models, Stable Diffusion, LLaVa and Vision Transformers in the front-end compiler. First ever proof of life of convolution-like operations on our compiler
• Developed operations & transformations for LLaMa2 on top of LLVM MLIR & torch-MLIR project architecture
• Spearheaded integration of LLaMa3 Decoder and Full Model into our compiler achieving 17.5x speedup in runtime

Visit IBM →

Toronto, ON

Python, C++, PyTorch, CV, AWS SageMaker

• Developed scalable ML software architecture to automate object detection tasks using Python and PyTorch
• Implemented, pruned, & quantized Facebook Research's Faster R-CNN model for a facial analysis task, used by 9 million users/year, reduced manual verification hours by 75%, and saved $35 million USD costs annually
• Trained model on a CUDA GPU and produced accuracy of 99%, F1-score of 0.99, and model size reduced by 78%

Visit BlackBerry Limited →

Waterloo, ON

Python, TensorFlow, NLP, NoSQL, Docker, AWS S3, EC2

• Developed a log anomaly detection platform combining NLP, data pipelines, and Elasticsearch using Python
• Implemented NLP model and improved model accuracy from 60% to 91% and F1-score from 0.30 to 0.87
• Productionized two anomaly detection models (Autoencoder + Isolation Forest AND Transformer architecture based on Google's paper) and retraining pipeline using TensorFlow, Docker, AWS S3, SageMaker & EC2

Visit RBC →

Toronto, ON

Python, Django, SQL, Exchangelib

• Spearheaded a comprehensive dashboard with 7 data source system integrations using Python and Django
• Improved productivity by saving 600 hours/month by developing an automated mailing response system
• Integrated a cloud database using Exchangelib, automated scripts, Django models & SQL queries

Toronto, ON

Python, JavaScript/TypeScript, React, jQuery, Node.js, Selenium

• Contributed on the Creative Pod, developing an interactive iframe with Python, JavaScript/TypeScript, React, jQuery, Node.js, and Selenium — building features, fixing bugs, and writing unit tests in a test-driven, agile environment
• Transitioned the codebase from Sinon/Chai to Jest using the Jest-Extended library, resulting in 2x faster tests running independently in parallel across threads
• Optimized two repositories by replacing libraries with manually implemented algorithms, increasing runtime of numerous components by 2–8x

• Leading extensive research into advanced distributed ML algorithms (data parallelism, federated learning)
• Implementing a P2P architecture to enable the connection of devices across large-scale distributed networks