Vladimir Kotrovskiy
Verified Expert in Engineering
Machine Learning Engineer and Developer
Vladimir is an experienced ML engineer with completed cross-disciplinary projects in areas ranging from NLP and CV to fintech and medicine. He can successfully develop and maintain a product from ideation to production in close contact with the business and collaborative teams, find solutions to all emerging problems, and always aims for excellent results. Vladimir has extensive experience with all major ML frameworks, writes clean code, and builds scalable and maintainable solutions.
Portfolio
Experience
Availability
Preferred Environment
Docker, Google Cloud, Linux, Amazon Web Services (AWS), Kubernetes
The most amazing...
...thing I've built is a flexible voice assistant SDK, allowing clients to create all types of dialogs and interactions, ASR activation, QA, and Wiki integration.
Work Experience
Data Scientist for legal SaaS
Berry Appleman & Leiden - Dunasi
- Designed and developed a complex product for real time data gathering, processing and analysis for constantly updated time series data.
- Developed system architecture and several SQL and NoSQL databases.
- Developed a distributed Redis queue system with multiple instances and workers per instance for parsing working fully automatically and monitoring tools.
- Worked on whitepapers and multiple predictors, including time series predictors, analytical pipelines, and R&D.
- Developed several backend microservices for mobile and web applications.
- Worked with Databricks ETLs and pipelines and integration between databases <-> Databricks <-> back\front ends.
Data Scientist, NLP
LoyaltyLoop, LLC
- Developed several variants of topic modeling, including transformers, statistical-based, and some mixtures.
- Carried out sentiment analysis, emotion detections, and scoring models.
- Found long-term trends and time-related changes in customers' attitudes.
- Made visual representations of the clustering process and topics, e.g., word clouds.
- Developed custom quality metrics and statistics for clusters.
- Completed back-end development with AWS S3, Instances, DynamoDB, Lambdas, VPC, and other services.
Senior Machine Learning Engineer
WorldQuant
- Build knowledge graphs from multiple datasets, developing architecture of graphs.
- Developed and applied machine learning models to graphs, such as PyTorch Geometric.
- Developed graph scripts, algorithms, and parallelizations.
Machine Learning Engineer / Data Scientist
MySky
- Built an ML, CV, and NLP system for recognizing tabular data from invoices and other documents, with a pipeline able to fast-train models for a specific invoice type.
- Developed multiple NLP/ML predictive systems for intelligent document processing, including named-entity recognition (NER), invoices and costs classification, text summarization, missing text prediction, and fraud detection.
- Worked on AWS services and back-end development, including Textract, S3, Lambda, SNS, PostgreSQL, DynamoDB, CloudTrail, EventBus, and CloudWatch integrations.
Data Scientist
MTS AI
- Built several NLP parts, including the NER, of a multi-agent medical AI advice system designed to dynamically make predictions based on brief initial claims, available laboratory data, and a series of consecutive follow-up questions.
- Created parts of the advising system responsible for generating relevant questions based on the current and previous state and integrated it with knowledge graphs.
- Developed and trained several transformer-based classification models (BERT, GPT) for a patient's diagnosis classification.
- Oversaw the integration of computer vision models into the general pipeline as well as designing and training alternative encoders.
Data Scientist, NLP
Alan, Inc.
- Developed crucial parts of the voice-assistant SDK, including the dialog flow engine, domain-specific NER models, sentiment analysis, and web and mobile screen/states integration.
- Created a full pipeline for automated NER training—from MTurk data gathering to production, in-training boosting, and a custom tool and templates for dialog creation.
- Implemented several alternative intents matching models, optimizations, BERT, and a custom CRF-based classifier.
- Designed and developed all of the small-footprint keyword spotting models used for voice activation and stoppage of customer applications (e.g., "OK Google"). It's integrated into existing applications.
- Developed a Wikidata integration used to answer general questions and another QA model created as an option for certain cases.
- Worked on a custom user-scripting language for customers to easily define their dialogs and adjustable ways to control the flow for a better experience.
- Added production monitoring, gathered cases and statistics, and automated NER models re-training based on such.
Data Scientist, Computer Vision
SmaSS Technologies
- Developed a face-recognition product capable of working in various surroundings, light conditions, and people moving and wearing some occlusive objects.
- Created high-speed algorithms used to select the most promising frames from the stream; also optimized and adjusted 3D ConvNets for the same purposes.
- Oversaw and was responsible for 3D face reconstruction (mostly SFS).
- Developed a search-and-comparison routine for reconstructed models.
- Fulfilled optimization/compression for small devices.
Data Scientist
First Moscow Medical University, Novartis
- Worked on a very collaborative medical project and was responsible for the overall architecture, design, and execution of R&D and the interactions with stakeholders and colleagues from diverse pharmaceutical companies and the university.
- Developed ML models for predicting a patient's response to oncological “targeted drugs,” depending on genetic, clinical, and other pieces of information—classical ML (XGBoost and AdaBoost) and many experiments/research with DNNs architectures.
- Implemented patient clusterization and searches for groups of patients related to the response based on genetic expression arrays, previous therapy, etc.
- Developed models that evaluated a patient's clinical status during treatment.
- Executed a search for possible interconnections between different physiological metrics, specific genes and the success of drug treatment, and disease progression with a more statistical-based approach (Gaussians mixture models, LDA, LSH, SVD).
- Defined novel training metrics describing a successful treatment based on patient and drug data.
- Completed the ML part in mostly standalone research on obstetrics—its correlations with genes, cellular pathways, nutrition, and other pieces of data.
- Developed computer vision (classification and object detection) models for CT and x-ray scans in several medical taxonomies and applied CNN for microarray scans for cancer classification.
Research Scientist
Keldysh Institute of Applied Mathematics (Russian Academy of Sciences)
- Developed and supported parts of a type of complex software used to calculate radiation-induced processes in layers of materials and various objects.
- Developed intricate algorithms in close collaboration with mathematicians, statisticians, and physicists.
- Completed the parallelization of computations and algorithms.
- Created a GUI on the Qt framework for some existing applications and added new features and ways of interaction with users.
Java Developer
Prime (previously Prime-Tass)
- Developed a client-server application for trading.
- Parallelized existing scripts. Developed general user interface in Swift.
- Analyzed and developed algorithms used in the company's software.
- Developed a program for traffic analysis in real-time.
Experience
ML Invoice Table Recognition System
http://mysky.comVoice Assistant SDK
http://alan.appMedical Diagnostic AI System
3D Face Recognition System
• Used video streams from CCTV cameras with overlapping fields of view.
• Applied different OpenCV and our own algorithms, partly based on 3D ConvNets, singling out several promising frames from different cameras, then refined and augmented.
• Used NN models, fulfilling several complete 3D-face reconstructions (mostly 3DMM and SFS).
• Maintained statistical metrics later used to merge these 3D models matched against an existing database.
Visa Processing Times Software
Skills
Languages
Python, SQL, Bash, C++, Octave, Java, R, JavaScript, Swift
Libraries/APIs
TensorFlow, PyTorch, SpaCy, Natural Language Toolkit (NLTK), Pandas, SciPy, Matplotlib, OpenCV, Scikit-learn, NumPy, XGBoost, Keras, Redis Queue, PySpark
Tools
Rasa.ai, Git, TensorBoard, Amazon Simple Queue Service (SQS), Amazon SageMaker, MATLAB, Gensim, GCC, Google Cloud Console, RabbitMQ, OpenAI Gym
Paradigms
Data Science, ETL
Platforms
Windows, Linux, Docker, Amazon Web Services (AWS), Apache Kafka, Kubernetes, Google Cloud Platform (GCP), Databricks, AWS Lambda, Azure
Industry Expertise
Healthcare
Storage
Data Pipelines, NoSQL, JSON, PostgreSQL, Neo4j, JanusGraph, Redis
Other
Machine Learning, Data Visualization, Computer Vision, Artificial Intelligence (AI), Natural Language Processing (NLP), Deep Learning, Podman, BERT, Data, Image Recognition, Object Detection, Data Engineering, Neural Networks, Chatbots, Statistical Analysis, Data Analytics, Startups, Time Series, Predictive Modeling, Business Models, Data Modeling, Image Processing, Convolutional Neural Networks (CNN), GPT, Parsers, Large Language Models (LLMs), Hugging Face, Generative AI, Distributed Systems, Quantitative Analysis, Transformers, Generative Pre-trained Transformers (GPT), Big Data, Topic Modeling, Architecture, DeepPavlov, Caffe2, Algorithmic Trading, Graphs, TigerGraph, Statistics, Clustering, Sentiment Analysis, Predictive Analytics, Data Mining, Analytics
Frameworks
Qt
Education
Completed a Non-degree Program in Bioinformatics
University of British Columbia - Vancouver, BC, Canada
Bachelor of Science Degree in Computer Science
Bauman Moscow State Technical University - Moscow, Russia
Professional Medical Diploma in Medicine
Russian State Medical University - Moscow, Russia
Certifications
Machine Learning
University of Washington
Machine Learning
Stanford (Coursera)
How to Work with Toptal
Toptal matches you directly with global industry experts from our network in hours—not weeks or months.
Share your needs
Choose your talent
Start your risk-free talent trial
Top talent is in high demand.
Start hiring