Created OkGPT, an AI personal assistant bot. It accepts a text, voice, or video message, transcribes it, and sends the result to GPT-4.

Integrated multiple APIs, including Telegram, OpenAI, Google, Redis, Amplitude, and more.

Optimized the code to parallelize user queries' processing to increase the bot performance by magnitude.

Supervised three team members and a few external collaborators.

Implemented support for most of the languages in text and voice messages.

Set up continuous integration/deployment procedures.

Developed an app that translated text in English to a special pseudo-phonetic alphabet.

Acted as a consultant to help define the deliverables and then the overall architecture of the translation tool.

Helped to define text annotation requirements and oversaw the annotation process.

Created a BI dashboard to Query and summarize a vast amount of semi-structured data.

Set up and tuned an Elasticsearch cluster and Kibana on AWS cloud.

Developed an ETL pipeline to ingest a terabyte of raw data into Elasticsearch.

Architected and directed the creation of a core Similarity engine to score candidates.

Created a Big Data pipeline in Databricks and Spark to enrich the input data and prepare the features for ML.

Used pre-trained NLP deep neural networks to create semantic text embeddings, which significantly increased the Similarity engine output results.

Developed a score using Spark GraphX to measure the company's attractiveness in the job market.

Prepared custom deep learning models to build richer embeddings, including various data sources and metadata.

Led communications with external data providers and created infrastructures to interface with their APIs.

Drove implementation of best DevOps and MLOps practices to improve reliability and reproducibility of ETL, feature generation, models' training, and inference subsystems.

Built a foundational end-to-end machine learning solution that predicts fair prices of real-estate properties, thus eliminating a need for manual assessment and enabling the company to run its business by providing quick responses to its customers.

Designed and implemented an automatically refreshing ETL pipeline that injects, cleans, joins, and enriches new data from AWS S3 storage daily.

Developed an interpretable machine learning model with Scikit-learn, CatBoost, Lifelines, FBProphet, FAISS, and SHAP that consists of several submodels and satisfies business monotonicity constraints.

Set up a continuous machine learning procedure for daily model retraining and redeploying based on the newly collected data.

Designed and implemented an automatic model promotion mechanism to ensure that models produced via the daily retraining process get deployed to production only if they have sufficiently good performance metrics and satisfy business constraints.

Created a historical data simulation system to generate synthetic data before the company’s launch and enable backtesting capabilities.

Architected and built the required infrastructure in AWS cloud: EC2 instances and VPCs, Docker environments for development, testing, and production, an Airflow pipeline for ETL and ML, and MLFlow model storage.

Created various dashboards for data exploration and data quality management, model performance monitoring, and visualized predictions.

Supervised other data science team members and coordinated with the engineering team.

Prepared and lectured a course on calculus for data scientists.

Developed and presented a course on linear algebra for data scientists.

Designed, prepared, and lectured a course on probability theory for data scientists.

Created a machine learning model that predicted revenues for a retail store chain based on store location, local demographic data, GIS features, seasonality, and other factors.

Developed and deployed an interpretable machine learning model that scored B2B customers for payment default risks and provided explanations for the scores. The model massively reduced workload for weekly risks assessment.

Built a probabilistic Bayesian machine learning model to predict which apartment buildings still under construction would fail to be commissioned in time. The model helped reduce the funds needed to hedge risks by two times.

Developed and deployed NLP models to automatically label a vast body of housing contracts by contract type and extract contractor party names, address entities, and other attributes.

Constructed and deployed a model to predict the problematic clogging of the evaporator in a chemical factory. This allowed for the timely preemptive service of the unit before it broke down, saving millions of dollars in production time.

Led and mentored a team of junior and middle data scientists in the projects mentioned above.

Communicated with clients, ensuring business goals were correctly translated into data science and machine learning tasks—explained insights and models to clients.

Architected ETL pipelines, including data acquisition, data ingestion, merging internal and external datasets, data cleaning and validation, data transformation, and feature engineering on several distinct projects.

Designed model performance metrics and their measurement protocols on several distinct projects.

Developed an ML system for a retail bank to recommend bank products to clients based on their past transactions' patterns. This included building an ETL pipeline and an ML recommendation system.

Invented a novel mathematical method for cross-frequency synchronization analysis in the human brain.

Implemented the method as a MATLAB toolbox and ran tests confirming that the results agreed with previously known scientific data.

Prepared and published the method and findings in a top-level journal.

Supervised the master's degree projects of several students.

Lectured a master's-level course on computational neuroscience.

Discovered a new geometric phenomenon accountable for the rigidity of certain mathematical models related to heat conduction in crystals.

Discovered a new stability property of attractors of multidimensional piecewise isometry maps related to Markov field models.

Discovered that almost every interval translation map of three intervals is finite type.

Prepared papers describing the findings and published them in high-ranking journals.

Created and verified mathematical models for the growth of blood vessels in the human placenta during the gestation period.

Developed the protocol for 3D data collection, including 3D surface scanning and micro-CT scans of the specimen.

Created an ETL pipeline to clean up and preprocess the collected samples.

Analyzed the collected data, fitted the mathematical models, and interpreted the findings.

Established that the rotation numbers of circle maps' semigroups define their generators.

Discovered a fractal structure of attractors of piecewise isometry maps related to Markov field models.

Prepared the papers describing the findings and published them in leading journals.

Lectured a PhD-level course on the structural stability of dynamical systems.

Discovered a new class of dynamical systems that have persistent massive attractors.

Established a deep relationship between skew product dynamical systems over Markov chains and nonlinear random walks.

Prepared the papers describing the findings and published them in major journals.

Designed and implemented algorithms for face detection on a 2D image, facial features detection, and alignment on a 3D surface.

Implemented and tuned 3D surface reconstruction algorithms.

Built and implemented statistical test procedures for new machine learning algorithms for face recognition.

Constructed and implemented a system for automatic test report generation based on log parsing.

Implemented and managed the automated build system dedicated to the building server.

Calibrated optical cameras and lasers for 3D scanners.

Migrated the algorithmic core to an embedded platform.

Developed a mathematical background for novel machine learning methods for handwritten text recognition.

Implemented novel methods for handwritten text recognition as a C++ library.

Presented my research at scientific conferences and seminars.

Developed an MVP of an app to query enterprise data in natural language. Given access to a database and a question in natural language about the data, the app would output the answer as a plot or a small table.

Engineered and fine-tuned the prompts to improve the quality and correctness of SQL code generation.

Created an automatic annotator for the database columns and the final table.

Created a machine learning model to automatically detect malicious smart contracts before they can cause harm.

Built a visualization tool for model output to audit its decisions.

Deployed the model to AWS cloud platform as a Lambda serverless function.

Reviewed and improved core courses in mathematics, data science, and machine learning.

Supervised the creation of new courses, including video lectures and exercises, on Data Science, Analytics, SQL, Power BI, and Tableau.

Recruited, interviewed, and screened lecturers and tutors for new courses.

Designed and implemented biometric machine learning face recognition algorithms.

Created and implemented statistical test procedures for new recognition algorithms.

Developed calibration procedures from 3D laser and flash scanners.

Refactored Windows32 code to make it cross-platform.

Implemented and tuned 3D surface reconstruction algorithms.

Applied software for barcode encoding and scanning.

Managed the build server and was responsible for CI/CD process for our team.