Evgeny Pavlov

LinkedIn, GitHub

Hi, I’m Evgeny! Currently, I work at Mozilla as a Staff Machine Learning Engineer. My main area of interest is prototyping and deploying ML/AI products in a full-stack manner – from Jupyter notebooks, ETL and training pipelines to scalable production services. I have experience shipping Deep Learning and Classical ML projects, including NLP/NLU, Machine Translation, Recommender Systems and Computer Vision in various companies. I’m based in Vancouver, Canada.

Current ML/AI projects

  • Firefox Translations (2021 – Present, Mozilla) – efficient on-device Neural Machine Translation in a browser (this work was initially a part of the EU-funded project Bergamot). I developed a scalable end-to-end training pipeline in collaboration with researchers from the University of Edinburgh, configured training infrastructure and trained translation models for 25 languages. Participated in development of Firefox Translations web extension. Currently leading the work on further scaling the training pipeline and improving quality of translations. (python, bash, C++, Slurm, GCP, Snakemake, Taskcluster, JavaScript, Weights and Biases)

Past ML/AI projects

  • Gather personal AI assistant (Gather, 2023) An AI assistant to analyze your private data (Python, OpenAI API, Llama 2, vLLM, FastChat, GCP)
  • Pocket recommendations (Mozilla, 2022) – Pocket similar content and “For you” recommendations (Python, Gensim, Lightfm, Metaflow, AWS, Qdrant)
  • Firefox Addon Recommender (Mozilla, 2021) – recommends Firefox addons on Firefox Settings tab and addons.mozilla.org (related add-ons). Migration to GCP, maintenance, ETL pipelines, analytics (Python, GCP, scikit-learn, Spark, Airflow)
  • Press Reader Deep PDF (2020) – Deep Learning based segmentation and block classification for PDFs of newspapers and magazines (Pytorch, Kubeflow, GCP)
  • Press Reader recommender (2019) – Deep Learning news recommender (Tensorflow, Python, k8s, GCP)
  • Native AI recommender (2019) – Deep Learning based “Read more” recommender for third-party websites (Tensorflow, Python, k8s, GCP)
  • News360 recommender (2018) – personalized news feed. Deep Learning based recommender, training pipelines, training infrastructure, a subsystem for AB experiments (Tensorflow, Python, Airflow, k8s)
  • News360 NLP platform (2017) – designed and implemented microservices to extract topics and entities from news articles (batch processing, inference of NLP ML models, Python, XGBoost).
  • Cognitive Forms (2008-2011) – OCR for structured documents. I developed an algorithm for multi-field contextual recognition as my diploma project. (C++)

Engineering

  • News360 (2016) – news crawler, analytics, migration of backend services from Windows C# stack to Python Linux stack using microservices architecture
  • Native AI analytics (2017) – monetization backend, analytics pipelines on Spark
  • NPTV (2013-2015) – C# framework for interactive TV apps with server-side rendering
  • Trud Expert (2011-2013) – desktop C# applicaiton on WPF

Pet projects

  • Attractions recommender – personalized recommendations of cool places to visit in British Columbia based on TripAdvisor reviews (Python, Pandas, Dask, Spacy, Gensim, ElasticSearch, React.js)
  • redhairtravel.com – travel blog on WordPress

Education

Artificial Intelligence Program, Stanford, 2021-2023

Master’s Degree in Applied Mathematics, MISIS, 2006-2011

Courses

  • Natural Language Understanding (Stanford Center for Professional Development, cs224u, 2023)
  • Machine Learning with Graphs (Stanford Center for Professional Development, cs224w 2022)
  • Natural Language Processing with Deep Learning (Stanford Center for Professional Development, cs224n, 2021)
  • Deep Learning specialization (deeplearning.ai, Coursera, 2018)
  • Machine Learning (Andrew Ng, Coursera, 2017)

Publications and talks

Hobbies

  • Music (electric guitar, drums)
  • Traveling (40+ countries)
  • Hiking, tennis, sailing, biking and other random outdoor stuff