Dmytro Ihnatov

• Led research and development initiatives, with a primary emphasis on Large Language Models and direct business application.
• Fine-tuned LLAMA2/Mistral base model to extract answers from the text and improved model performance from 24% to 72% of Exact Match metric.
• Applied GPT-4 API for automatic labeling of the data.
• Developed the entity resolution model.

• Provided consultations for private companies regarding AI applications.
• Developed and wrote a comprehensive strategy for AI integration into products, tailored to the specific business objectives of each client, resulting in improved efficiency and innovation.
• Led the research and development efforts of Large Language Models (LLMs) for direct business application.
• Developed an instruction-based text summarization model.

• Implemented item embedding model and merged it with the internal model collection.
• Improved the quality of video recommendations by connecting two new models to the service backend.
• Prepared and initiated A/B tests with a team to validate model performance before production.
• Presented a new recommendation model and concept to the team during the paper review.

• Built machine learning product for investment fund managers to uplift sales
• Scraped data from 10+ financial institutions and social media
• Built the NLU model to extract a signal of specific interest from text

• Worked on testing and improvement of a recommender system of highly trafficked websites

• Conducted research in the domain of Recommender systems
• Designed the custom RecSys model based on the implicit user feedback
• Lead the development of RecSys models for the machine learning platform (Kepler)
• Contributed to the NLU solution for a client project by developing a custom metric using NLP toolkit

Accomplished a list of academic and industrial projects described below.
Shoeby, Netscalers (Doctorate Thesis) | September 2017 – August 2018

• Led the project and reported results to CEOs of two companies
• Translated data insights into business recommendations
• Developed a hybrid end to end recommender system in Tensorflow with visual and return ratings to increase net sales of the webshop
• Used InceptionV3 model to extract visual features
• Built a web app (Dash\Flask) to visualize recommendations and additional insights for the data team

Tilburg University (Researcher | Business analytics module) | April – June 2017

• Participated in the development of linear and heuristic optimization models using python and GUROBI package to optimize the benefit of proton radiotherapy facility over multiple patients
• Built a Graphical User Interface in Bokeh to visualize the result of optimization according to selected parameters

MyHomeServices (Data Engineer) | February – April 2017

• Designed a new SQL schema of the production database and document-oriented schema in the MongoDB
• Evaluated the performance of the current production database and two new schemas

Hospital (Jeroen Bosch Ziekenhuis) (Data Scientist) | December 2016 – February 2017

• Participated in the project of early-stage identification of people with Primary Antibody Deficiency
• Applied several machine learning technics (Random Forest, Decision Tree, SVC, LDA, K-Means) to select features/symptoms that lead to disease detection

Eindhoven University of Technology (Researcher | Data mining module) | October – November 2016

• Built CNN and DNN models in Tensorflow for digit classification (MNIST dataset)
• Extracted features from images for DNN by applying several algorithms (Canny, Corner Harris, Structure Tensor, HOG)
• Implemented the beam search algorithm to find regions of the search space where DNN does not perform well

Rockpanel (Data Scientist) | September – December 2016

• Analyzed sensor data of the production line
• Applied machine learning technics to identify key factors that lead to bad quality of products