Developed the back end for the full ML cycle, ModelOps, and MLOps, on the platform. Added a wrapper on top of the AWS SageMaker.

Worked on dozens of automation workflows (use cases), including MLOps, analytics, DataOps, networks, ITOps, etc.

Created the back- and front-end elements using React for a drag-and-drop, chatbot-building application.

Implemented and deployed dozens of algorithms (classification, clustering, time series, NLP, and computer vision) for different use cases.

Led a small ML team, including planning, management, monitoring, and leadership.

Sr. Data Engineer | Grata | Developed async ML based distributed pipeline for website pages classification, data scraping, locations extraction and normalization.

Sr. AI Engineer | Briefly |Owned BE, ML and DevOps parts of the project, applied LLMs to build productivity tool features (summarization, fine-tuning, RAG, recommendations, embeddings, inf. retrieval, etc.). Slack App, API integrations, data engineering.

Sr. AI Engineer | Wing | Technical lead of the team. Analytical engine implementation, agentic chat bot for financial advisory implementation, RAGBtesting.

Developed an ingestion and processing pipeline on AWS for photos from surveillance cameras.

Experimented with various non-DL and DL approaches and tested them. Trained and used the Siamese network with an attention mechanism, achieving +95% accuracy.

Created and shared presentations with analytics to executives. Developed and maintained a wiki for the project.

Led and mentored a team of students. I defined objectives and controlled the process using the Agile methodology.

Created a tool for crop classification and map creation.

Collected data manually and via web scraping using Mapillary and oversaw data labeling.

Implemented and tested DeblurGAN and several other classic deblurring methods.

Oversaw field localization (YOLO) and crop classification by fine-tuning a ResNet model.

Created a multi-staged data pipeline from raw packet data (TCP/IP layer) to consumable inputs for machine learning models with multi-processing implementation in Python (CPython).

Trained, tuned, evaluated, and compared multiple machine learning models in Python (scikit-learn, Keras, XGBoost, CatBoost) and C++ (mlpack).

Oversaw the data analysis and communication with stakeholders. Created a reusable Python tool for generating rapid and externally configurable data analysis reports.

Implemented custom feature generation algorithms based on expert knowledge based on aggregation, derivatives, TCP/IP conversation delays, products, and fractions.

Implemented custom, multi-staged feature selection algorithms that were model-based.

Deployed and monitored the project in production in the network. If the tool detects congestion, optimization policies are applied. Customers reported up to a 20% increase in the quality of delivery for video content.

Developed, deployed, and evaluated a hybrid recommendation engine in Python for an online store.

Oversaw customer behavior analysis, visualization, and stakeholder communication.

Handled different scenarios of user engagement using a strategy pattern. Contextual recommendations were given based on popularity (general and category-based), item-to-item, and SVD. (Python, scikit-learn, SciPy).

Integrated recommendation engine into a Django back end.

Conducted multiple A/B tests with random sampling for evaluation of the system. Compared to the most popular items in the category baseline, we achieved up to a 150% boost in purchases per session and increased revenue.

Created an image super-resolution module for an online cloud site constructor with Keras.

Utilized middle-deep CNN, trained on several blur kernels, and deployed it as a service via REST.

Conducted surveys showing an increase of about 5% in satisfaction for users of the platform.

Developed algorithms for smart keyboard functionality (word prediction and spelling correction).

Developed Naive Bayes for n-grams and K-nearest neighbors (KNN) for spelling corrections.

Created tweaks for better algorithm performance using Laplace smoothing and a custom keyboard distance for KNN.

Developed algorithms with C++. Integrated them with a Java to Android keyboard and published them to the App Store.

Developed a Content Management Interoperability System in Jakarta EE. I used the Model-view-controller framework for the application.

Developed Enterprise JavaBeans for the business logic of the application.

Developed JavaServer Pages for the presentation level.

Co-developed multiple Django-based services on the back end for data processing.

Utilized and adopted OpenAI models for multiple tasks, such as summarization, classification, and recommendations.

Performed APIs integrations of external services and tools.

Worked on AWS development using AWS Lambda, Amazon SQS, Amazon DynamoDB, and Amazon SES.

Developed end-to-end distributed NLP-based geocoding pipeline using Celery on Kubernetes.

Implemented scraping from company websites and aggregators.

Developed and deployed on the AWS SageMaker web page category classification model.

Implemented a hybrid geocoding model with query lookup, query relaxation, result validation, prioritization, and fallback mechanisms.

Used combinations of available geo databases and offline entity extractors like libpostal and third-party geocoding services to combine it in a single view.

Increased a fraction of parsed addresses, reduced incorrect addresses by 95%, and improved the overall data quality score by 20%.

Performed a time-series forecasting of the demand for e-scooters for a global e-scooter rental company (hundreds of cities) with models for auto selection and auto retraining.

Performed ad-hoc data analysis and built Looker dashboards.

Performed an intervention-effect analysis (for promotions and other events).

Introduced and implemented MLOps techniques, tools, and approaches.

Rebuilt a dozen monolithic R pipelines into distributed, modular, and functional-styled Python pipelines.

Developed MLOps layer on top of Dagster, Seldon, Feast, and other tools.

Fine-tuned existing model hyperparameters both for speed and performance.

Introduced MLOps tools to existing infrastructure (Seldon, Feast, and Great Expectations).

Migrated existing hardcoded models to introduce the MLOps infrastructure.

Developed back-end APIs using Django for ML-related services.

Implemented classification and time series forecasting models for manufacturing sensors.

Trained and evaluated AdaBoost models for customer churn prediction using FBLearner Flow.

Performed hyper-parameters tuning for optimization.

Data-engineered with Hive and processed data using Python.