Yuriy Sosov

AI ML RL Data Science PROJECTS: >> Senior Data Scientist AI/ML, various clients/projects, remote Apr 2024 - present Scope : • AI assisted IT development: requirement validation, code generation, code debugging • developed: validation of GPT output, GPT-RAG, OCR, recommender-systems, MCTS - RL ( Monte Carlo Tree Search - Reinforcement Learning ) simulations for logistics optimization >> Senior Data Scientist for Generative AI @ Helasoft Jul 2023 - Apr 2024 via MBIT • Reviewed the current BERT implementatios for text extraction; made error analysis • proposed improvement • Reviewed LLAMA-2 vs Open AI GPT vs BERTs for text extraction • (in a team) integration of Open AI GPT and LLAMA-2 • modified LLaMA-2 generation • developed and deployed (modified) LLaMA-2 7/13/70B (Flask) server endpoint at AWS SageMaker EC2 • developed prompt and dialog logic for text extraction with OpenAI GPT and LLaMA-2 • fine-tuning gpt-3.5-turbo on gpt-4 corrected output Skills, Tools and Technologies Used: pytorch, HuggingFace transformers: BERT, LLaMA-2, OpenAI api >> Senior Data Scientist @ SignUp Software Jul 2019 - Apr 2023 via Developers Shore Scope: Developed ML system for automatic mapping of invoice data into transactions of Microsoft Dynamics 365 Accounting. The trained models run as json endpoint in Azure Kubernetes or Docker Containers • developed ML model / regression with proprietary mapping, custom stochastic descent with constraints, Munkres assignment • developed ML model Azure deployment script Tools and Technologies Used: • python: Pandas, NumPy, SciPy, Numba, Matplotlib, Plotly, clustering, embedding, Scikit-Learn; PyTesseract; • LLM: BERT, GPT • cloud Azure ML API: Experiments in Compute Instances / Clusters, Endpoints in Kubernetes / Docker

Scope: Analysis and porting of over 20 large applications accessing the Sybase/Oracle to SQL Server • Analyzed interfaces of 8 public libraries [C/C++: db-lib, ct-lib, roguewave (DB Interface Module) perl: db-lib, ct-lib, sybperl, dbi java: jdbc] and 7 Barclays’ developed applications (utilizing c/c++/perl/java/c#/shell) for the code accessing the Sybase/Oracle databases • Ported access methods of several applications from Sybase/Oracle to SQL Server Tools, Technologies and Skills Used: core C / C++, core java, perl, shell scripting, unix, MS Sharepoint, libs for db access( C/C++: db-lib, ct-lib, roguewave (DB Interface Module) perl: db-lib, ct-lib, sybperl, dbi java: jdbc, c#: odbc; )

Scope: Analyzed performance of the data-mart / datawarehouse ETL and reporting procedures. Provided optimization solutions. Developed and modified db packages used in ETL and web reports. Responsibilities: • Supporting the business side in business requirement gathering and specification analysis within the assigned projects • Design, development and optimization of the database ETL and reporting procedures (also developed an etl performance monitoring tool in vb-excel) • Refactoring and performance tuning of the existing database software, query tuning, index optimization, partitioning and replacement of the large and frequently used tables • Development of reliable performance testing routines, statistical analysis of log files, spotting performance trends, performance prediction and estimation • Unit testing • Creating deployment scripts • Troubleshooting of production issues Tools and Technologies Used: Oracle 10g.2, PL/SQL, SQL*PLUS, SQL*LOADER, SQL Navigator, MS Source Safe, vb6 excel

Scope: Provided integration upgrade of three ETL (extraction transformation loading) systems to be compatible with the consolidated CTMS (clinical trial management system) database and the upgraded version of IMPACT (data input / output / reporting system) Responsibilities: • Modified db packages to be compatible with the upgraded version of IMPACT • Modified the VB client for one loader • Modified the J2EE component of one loader using Eclipse IDE • Modified Informatica mapping components • Resolved two emergency cases and a UNIX privilege issue in the production environment Tools and Technologies Used: Oracle 9i/10g.2, PL/SQL, SQL*PLUS, SQL*LOADER, UNIX Shell Scripting, windows command batch scripting, J2SE/J2EE, VB/VBA, JDBC/ODBC, database modeling, XML, MS Visio, Eclipse, MS Visual Studio, Informatica, ftp, telnet, vi

• Developed a parallel finite-difference solver for the time dependent Boltzmann equation expanded into Legendre polynomials • Modified a 6-dimentional Monte-Carlo solver for the time dependent Boltzmann equation • Developed a finite-difference solver for Drift-Diffusion equation • Developed an Over-relaxation solver for the Poisson equation • Developed a Solver for 2D electric current distribution in solar cells • Deployed the above applications at the P4 Xeon Infiniband Cluster of Ohio Supercomputer Center • Developed applications for input/output, statistical analysis and graphical representation of the data for the above projects Tools and Technologies Used: Mathematica, C++ / C / FORTRAN under Linux / Windows, Parallel Programming in Computer Clusters with MPI, Bourne shell scripting

(SIAR-Bossard was acquired by CapGemini in 1997, and was merged with the CapGemini subsidiary Gemini Consulting) Responsibilities: market entry and acquisition studies, support of market entry decision and acquisition projects in Ukraine for the Scandinavian and International clients: McDonalds Europe, Baltic Beverages Holding, Swedish Match, Girobank, Vin and Spirit, Primalco