Armin Norouzi

Armin Norouzi

Applied AI Software Engineer at Google

Work Experience

Software Engineer, Aplied AI (December 2024 – present)

Google, Waterloo, Canada

  • Worked as an Applied AI Software Engineer in the Google Cloud Platform (GCP) Contact Center as a Service (CCAI) Platform.

Software Engineer, Machine Learning Engineer (December 2022 – November 2024)

The Citco Group Limited, Toronto, Canada

  • Architected and led the Citco bank transaction settlement dates prediction and total deposit balance forecasting, managing the code base and developing and deploying multiple machine learning and deep learning models.
  • Accelerated ML development lifecycle by 25% by leading the development of a custom Python package incorporating Spark-based ETL pipelines, the model pipeline with continuous training, registering, deploying, and experiment tracking.
  • Designed and implemented five ML pipelines for different transaction categories, leveraging custom-built AutoML for feature and model selection, achieving a consistent 9% accuracy improvement over baseline models and preventing model drift.
  • Developed and released a deep learning model using multi-tower architecture and factorization machine techniques to forecast bank deposits, resulting in an average $150 million increase in weekly balance forecast accuracy compared to the legacy model.
  • Developed and deployed LLM-based extraction techniques leveraging serverless AWS infrastructure to automate key data extraction from trade documents. Achieved an 18% improvement in extraction accuracy compared to the legacy system, resulting in annual cost savings of over $1.2 million.
  • Developed and deployed an AI chatbot with semantic knowledge search using Databricks vector search and DynamoDB caching, achieving over 85% user satisfaction.
  • Ensured optimal performance and reliability of the CitcoRecs transaction reconciliation system by providing ongoing support and maintenance for over 20+ ML models deployed on SageMaker.

Technical Writer (AI and computer science) (May 2022 – November 2024)

Medium, Freelance

  • Published a comprehensive series on data structure and algorithm series and ML system design series, elucidating over 100 algorithms using Python. The series offers profound insights into the architecture of significant large-scale ML systems.
  • Authored an in-depth series on Generative AI and Machine learning and data science, comprising over 20 articles, emphasizing LLM fundamentals, including topics like transformers and adversarial networks, supplemented with Python code.
  • Published articles in various platforms, including esteemed medium journals such as Better Programming, TowardsAI, Level Up Coding, Lakers.ai blog, and my blog showcasing advanced topics in LLMs and AI.

Lead Machine Learning Engineer (August 2023 – November 2024)

MA EdTech Solutions Inc, Toronto, Canada (Part-time)

  • Developed innovative AI techniques for analyzing children’s diary entries called parentzapp.com, cross-referencing them with psychological literature using vector search and Retrieval-Augmented Generation (RAG) techniques. This initiative aimed at the early detection of mental health issues and potential abuse.
  • Led the development of AI backend using serverless REST API on AWS, integrating services such as API Gateway, Lambda, AWS Bedrock, Databricks, and Pinecone VectorDB with focusing on creating scalable machine learning architectures suitable for production and robust deployment strategies.
  • Led the implementation and management of Continuous Integration and Continuous Delivery (CI/CD) processes utilizing GitHub, AWS CloudFormation, CodePipeline, and CodeBuild and guided a technical team to ensure streamlined development cycles and the delivery of high-quality software.
  • Streamlined AI-driven backend functionalities by orchestrating front-end CI/CD configurations via AWS Amplify in collaboration with front-end engineering teams.

Software Engineer, Data Scientist (June 2022 – December 2022)

Skyjack Inc, Guelph, Canada

  • led a team of interns and collaborated with R&D, service, and marketing to design and implement ML pipelines, resulting in the successful delivery of 3 projects within 6 months.
  • Created and productized ML model using AWS CodePipeline, SageMaker, and Lambda for predicting machine battery levels with 91% accuracy based on telematic data streamed in real-time from Skyjack machines globally.
  • Optimized engineering and product team efficiency by conducting in-depth analytics, presenting findings through visualizations using Plotly through an AWS web application, and creating automated reports five times faster than manual ones.

Doctoral Researcher and Lecturer, Machine Learning and Intelligent Control Systems (May 2018 – May 2022)

University of Alberta, Edmonton, Canada

  • Designed end-to-end ML solution using a recurrent deep neural network in TensorFlow, achieving a 96% accuracy in predicting combustion engine emissions. This innovation eliminated the need for high-cost sensors and expedited the design and implementation of the controller, making it 10 times faster than traditional approaches.
  • Led research teams in three major industrial projects, replacing traditional automotive system modelling with accurate, data-driven ML solutions in collaboration with partners like MMP Lab, IAV GmbH, and Cummins.
  • Created a comprehensive Machine Learning course (MECE 610), covering essentials to advanced ML and deep learning topics, supplemented by over 50 practical Python examples, both from scratch and using standard packages.
  • Executed and facilitated two highly engaging workshops on machine learning and computer vision, providing critical insights to 120+ attendees at the CSME conference and the University of Alberta.

Research and Teaching Assistance (September 2015 – February 2017)

K.N. Toosi University of Technology, Tehran, Iran

  • Developed a novel method for autonomous vehicle control using computer vision, meta-heuristic optimization, control theory, and fuzzy logic systems and deployed via MATLAB/Simulink and CarSim co-simulation, significantly increased the vehicle’s stability and comfort during various maneuvers, reducing lane keeping and lane change tracking error by 40%.
  • Provided supervision for an undergraduate capstone project, focusing on integrating control theory and meta-heuristic optimization within autonomous vehicle control.
  • Served as a Teaching Assistant, leading graduate course classes on dynamics, vibration, and engineering mathematics, ensuring a comprehensive understanding of the subjects.
  • Authored and shared a detailed tutorial for 20-sim software, enabling users to develop multi-domain dynamics models using the Bond graph method, enhancing user proficiency and model accuracy.