Maria – Python, MySQL, PostgreSQL

Contribute to the automated lead bidding system for the auction. Investigate the unexpected poor performance of bidding models. Create new models for targeted marketing. Research and process large-scale table data to implement new features and improve performance for newly updated datasets. Re-create modelling and data processing pipelines.

• Analyse and investigate poor model performance
• Create local databases using Docker
• Participate in database migration to Azure
• Form and filter datasets for training
• Develop and train new models
• Optimize codebase for training and evaluation processes
• Research and process new features to implement for modelling
• Re-create pipelines via Apache Spark

Achievements:

• Improved model performance by identifying root causes of degradation and redesigning training datasets and feature pipelines, leading to measurable gains in accuracy and stability.
• Migrated infrastructure to Azure and rebuilt scalable data pipelines with Apache Spark, significantly accelerating training and evaluation cycles while optimizing the overall ML workflow and deployment reliability.

Create a platform for scanning products, classifying them, detecting specific features, and generating marketing descriptions.

• Classify products on the images and data based on the ontology provided by the customer
• Extract labels, fit, material composition of product, etc
• Design database and data structure
• Generate descriptions for products with OpenAI in German
• Create pre- and post-processing modules for asynchronous generations, processing, and filtering systems
• Logging and solving bugs in the service

Achievements:

• Increased product data accuracy and reduced manual catalog processing time by automating classification, attribute extraction, and multilingual description generation, enabling faster time-to-market for new items.
• Improved content consistency and system scalability through intelligent filtering, ensuring reliable large-volume product enrichment.

Developing a chatbot for polling a huge number of people regarding their political preferences for a particular event in Australia.

• Manage the project and lead the team
• Design conversational flow for pooling
• Build the conversational flow
• Implement various types of Questions for different expected or non-expected human behaviour
• Apply Fallbacks as a part of conversational flow for questions without expected replies
• Tweak the settings of the agent to imitate a real human agent
• Manage demos, big groups of unaware humans, tests, and upgrade the flow based on results

Achievements:

• Led the end-to-end development of a conversational polling agent, designing and optimizing dialogue flows that increased user engagement and completion rates across large-scale live test groups.
• Improved interaction quality by implementing adaptive question logic, smart fallbacks, and human-like agent tuning, iterating the system based on demo feedback and real-user behavior insights.

Creating from scratch of fully automated service for various predictions of cities in OAE load based on current events taking place, weather, and historical data from multiple non-synchronised sources

• Manage the project and lead the team
• Develop parsing modules for all the data sources and analyse the data
• Build time series forecasting models based on historical data
• Design and create a database for storing and continuously updating data
• Process geographical data for covering the whole area of different geographical levels of interest (country/emirate/city/district/etc)
• Build pipelines for asynchronous runs for different models
• Logging and solving bugs in the service

Achievements:

• Led the delivery of a data-driven forecasting platform by coordinating the team and building scalable pipelines that automated multi-source data ingestion and continuous database updates.
• Developed time series models and geospatial data processing covering multiple geographic levels, enabling accurate regional forecasting and ensuring high system reliability through robust asynchronous workflows and proactive issue resolution.

System of artificial intelligence models and algorithms for generating contextual selling advertising headlines and descriptions for texts scraped on websites in Danish and English.

• Manage the project and lead the team
• Develop a system for generating text based on descriptions from a website using Markov Chains and LLMs
• Improve service with a filtering system for generations based on distance metrics like cosine similarity, Levenshtein distance, Fuzzy logic, etc
• Include various types of analysis like sentiment, polarity, linguistic analysis, etc
• Implement an algorithm for keyword extraction
• Create and implement an algorithm for generations based on templates
• Add a system for human-in-the-loop evaluation for constant improvement of project models
• Logging and solving bugs in the service

Achievements:

• Led the end-to-end development of an AI-powered text generation system, Markov Chains + LLMs, introducing advanced filtering with cosine similarity, Levenshtein distance, and fuzzy logic, which significantly improved output relevance and reduced low-quality generations.

Development of a fully automated lead generation system to find more qualified leads, place them in the sales funnel, and ultimately generate more quarterly revenue for the company.

• Manage the project and lead the team
• Analyse and filter the dataset
• Build a few custom text classifiers to analyse human speech for sales management
• Implement various methods for context and text analysis, like sentiment analysis, language detection, part-of-speech analysis, etc
• Design and build the database structure
• Logging and solving bugs in the service

Achievements:

• Led a cross-functional team to successfully deliver a speech analytics solution for sales management, improving call evaluation efficiency and enabling data-driven performance insights.
• Built and deployed custom NLP classifiers and optimized the database architecture, increasing text analysis accuracy and enhancing system stability through structured logging and proactive bug resolution.

Cell distortion recognition in clinical trials. A Platform to speed up the process of searching for and selecting high-quality sperm, while maintaining a high accuracy to create more targeted treatment options, to give people who are fighting for the possibility of conception a greater chance of success.

• Create an algorithm to determine spermatozoids in the video feed from the microscope camera
• Classify items according to various mutations to choose the most suitable samples for conception
• Detect the position, movement, direction and speed of movement
• Upgrade the processes by applying filters for better tracking
• Logging and solving bugs in the service

Achievements: Upon successful completion, I delivered an artificial intelligence system trained on a diverse dataset of male fertility samples, capable of analyzing microscope-captured images, quantifying sperm cells, and assessing their motility, morphology, and concentration.

The goal of the project is automate the processing of unstructured financial trading documents and obtain the information necessary for conducting financial transactions in order to increase the throughput, turnaround, and thoroughness of trade operations.

• Extract and identify key fields from documents to classify their type and predict relevance for subsequent analysis
• Develop new features, algorithms, models, and update existing ones
• Improve the evaluation process and invent metrics for business purposes
• Logging and solving bugs in the service
• Create a demo for extracting information in real-time, only in the highlighted area with a click
• Constantly improving existing features and processes

Achievements:

• Increased accuracy of extraction over 80%
• Invented a business metric for the report
• Invented a demo for extraction from a single page with accuracy ~70%