Oleksiy – Python, Tensorflow, Keras

The recommendation system designed to recommend essential oil blends being based on data from Apple Health and user’s answers to regular surveys and dynamic questionaries. Evaluation of the user's emotional state based on the obtained data allows to manage personalized recommendations of oils rituals to focus on achieving calmness, improving sleep, restoring energy, or helping with centration.

• Generated rituals and oils descriptions, tags, and other characteristics;
• Developed the logic to detect stress and sleep quality based on apple health data;
• Designed surveys and questionnaires to get initial and additional information about customer health state and activities;
• Designed and implemented the recommendation engine based on collected data;
• Developed and dockerized a REST API endpoint for the recommendation system;
• Deployed dockerized solution on AWS;
• Prepared and managed presentations;

A machine learning question-answering solution tailored for a fantasy game, functioning as an interactive chatbot equipped with an expansive understanding of the game's world. This allows players to engage with a persona that readily provides insights about the game universe. Beyond the retriever-reader elements, the solution encompasses grammar correction and paraphrasing functionalities to elevate the answer quality. Additionally, it seamlessly integrates with ChatGPT, ensuring the capacity to maintain contextual coherence across ongoing discussions

• Implemented the question-answering machine learning solution for the chatbot;
• Executed Extractive Q&A, incorporating grammar correction features;
• Conducted comparative analysis among pre-trained models: Bert family, Electra, Roberta, and Sentence transformer;
• Executed question paraphrasing techniques;
• Applied coreference resolution;
• Created a conversational bot through seamless integration with ChatGPT.

The tool to assist with analyzing product reviews. It is an ML solution based on fine-tuning GPT-3 for evaluation of evaluation of completeness, grammar, and relevancy of product reviews.

• Implemented solution for evaluation of completeness, grammar, and relevancy of product reviews;
• Developed API and simple web app to work with the solution.

The solution for generating the text summary and generating the most relevant keywords for unpublished web pages to hep running the SEO process.

• Implemented keywords extraction solution for unpublished web pages;
• Created text summarization for web content;
• Conducted training of Bert models to extract keywords;
• Developed GPT-3 models for keyword extraction purposes;
• Enhanced prompt engineering within GPT-4 to improve keyword extraction.

It assists real estate managers in developing the building's descriptions by automatically generating characteristics of real estate objects based on their images.

• Worked on WBS and estimation;
• Collected data;
• Discovered available solutions;
• Evaluated model;
• Prepared and managed presentations.

System to facilitate the process of extracting and validating data from various document sources, including but not limited to emails, dates, and organization names.

• Investigated models for OCR and NER;
• Applied Tesseract OCR;
• Evaluated the results;
• Tuned hyperparameters to improve performance;
• Prepared and managed presentations.

A comprehensive system designed for the processing of COVID test results, encompassing rigorous verification of the testing procedure, including facial recognition, test direction confirmation, detection of essential nasal elements, and the assessment of sweat presence within a person's nostrils. Additionally, the system incorporates an OCR (Optical Character Recognition) component, enabling seamless extraction of text data from uploaded documents and automatic population of the relevant fields within a registration form.

• Applied object detecting model;
• Developed customized guidance for the management of COVID tests;
• Applied image classification model;
• Prepared and managed presentations.

The system collects behavioral characteristics during the shopping trip of customers and predicts the probability of potential mistakes in the list of products scanned by the customer.

• Implemented the classification model for detecting customers who tend to fail to rescan;
• Improved the performance of the model by feature engineering;
• Deployed the solution to the cloud.

The system discovers association rules by applying statistical analysis of the probability of items to be purchased separately and the conditional probability of items bought together. To explore the most relevant associations, it exploits a priory algorithm based on support, lift, and confidence for each group of items.

• Implemented market basket analysis algorithms;
• Researched with the most relevant association rules;
• Presented results to business stakeholders.

The model is trained on customers' purchase history. Particularly, it considers Recency, Frequency, and Monetary metrics to segment customers into homogeneous groups. The discovered customer behavior patterns detect the need for engagement with relevant campaigns to prevent customer churn.

• Implemented RFM model for customer segmentation;
• Prepared presentations for stakeholders.

System for the enrichment of football translation in real-time by identifying the positions of the ball and players on the football field to manage the camera direction. Additionally, it includes the prediction of the ball's direction leveraging the Kalman Filter.

• Applied detection model with additional post-calculations;
• Implemented image augmentation;
• Investigated and applied additional algorithms, like the Kalman Filter;
• Executed various custom logic to improve the object due to specific requirements;
• Prepared presentations.