AI Engineer

Experience

Senior Machine Learning Engineer — Nov 2025 – Present

• Take part into the design and development of AI-powered solutions, including AI Assistant, AI Assessment, AI Playlist, and Content Authorization tools, using Python and any AI frameworks and LLM models.

• Migrated a Transcription service to Microsoft Azure, implementing scalable and reproducible infrastructure using Bicep (Infrastructure-as-Code), including Azure Functions, Azure Storage, and Azure Speech Services.

• Built and maintained separate infrastructure and application CI/CD pipelines in Azure DevOps, enabling automated deployments and environment parity across DEV, UAT, and PROD.

• Implemented Azure Function based integrations with Azure Service Bus, designing and configuring queues and topics to support asynchronous messaging.

Senior Machine Learning Engineer — May 2025 – Nov 2025

• Designed and delivered agentic solutions powered by ELITEA, improving scalability and efficiency across business processes.

• Organized and led AI workshops to raise awareness, develop skills, and drive integration of AI technologies across teams.

• Developed a Scrum Master Agent that converts user stories into structured Jira tasks for Backend and Frontend engineers, automatically generating acceptance criteria.

• Created an Agile Assist Agent that enables internal teams to quickly access project information, make informed decisions, and update Confluence pages as needed.

• Built a GitHub Agent to streamline onboarding for new engineers by analyzing codebases, generating documentation, and automating branch creation for new code changes.

• Implemented a CI/CD Orchestration Agent capable of analyzing pipelines, approving successful runs, and automatically fixing, errors when detected.

Machine Learning Engineer — Aug 2024 – May 2025

• Enhanced a scalable message processing pipeline with Speech-to-Text (STT) using Whisper, ASR for translation and diarization, and GPT models for transcript processing.

• Advancing multiprocessing in LLM operation management and enhanced RabbitMQ queue

• Adapted Zoom transcripts with advanced diarization techniques to accurately identify and attribute speech to defined speakers, achieving a higher level of precision and clarity in speaker segmentation.

• Annotated and prepared high-quality datasets for pyannote fine-tuning, enabling improved speaker diarization accuracy and performance for domain-specific audio transcripts.

• Streamlined and advanced prompt management by integrating Contentstack as a centralized system, migrating all prompts to enable real-time updates and improved flexibility, enhancing the efficiency of prompt iteration and deployment.

• Designed and executed prompt engineering experiments for Assessment Skill Evaluation tasks using LangChain, working with cutting-edge LLMs including Anthropic Claude Sonnet 3-7, DeepSeek-R1, Gemini 2.0 Flash-thinking, and Amazon Nova. Developed and iterated custom prompts to evaluate model performance across reasoning, consistency, and output quality dimensions.

• Designed and developed a recommendation system for a learning portal with a robust feedback loop to enhance personalized learning experiences.

• Built an ETL pipeline for seamless data extraction, transformation, and loading, integrating with BigQuery for analytics and Redis for efficient event data storage.

• Integrated the system with the Milvus vector database for similarity searches and implemented business rules to align recommendations with project requirements. Ensured scalability, real-time data processing, and improved system performance to deliver accurate and adaptive recommendations.

• Designed and implemented Apache Kafka topics with Avro schema definitions, developed a custom Kafka producer for data ingestion, and orchestrated the pipeline execution using Apache Airflow DAGs.

• Audited an AI-driven English Assessment portal to ensure results closely aligned with expert manual validation, enhancing accuracy and reliability of the system.

• Developed a comprehensive data processing pipeline that includes robust data preprocessing, feature extraction, and automated generation of performance metrics such as Precision, Recall, and F1-score.

• Conducted evaluations using two distinct prompt variations with state-of-the-art LLM models: GPT-4o-mini-2024-01-18, Gemini-1.5-pro, and Anthropic Claude-v3.5-sonnet. Analyzed and compared model performance across key metrics.

Machine Learning Engineer — Jun 2023 – Jul 2024

• Constructing and orchestrating a data processing pipeline, which involves the systematic design and management of data flow from initial collection to Machine Learning model training and final analysis.

• Creating evaluation metrics to systematically assess and measure performance. This process is essential for quantifying the effectiveness of various strategies and approaches, allowing for informed decision-making and continuous improvement.

• Developing unit tests and integration tests to ensure the robustness and reliability of the system, followed by constructing a Continuous Integration/Continuous Deployment (CI/CD) pipeline utilizing Docker to streamline and automate the deployment process

Data Scientist — Dec 2019 – Nov 2022

• Developing ML-Based Scoring Module which enables to offer loans at a variable interest rate trained the model using the Linear Regression algorithm, achieving an 84% R2 score.

• Developed Probability of Default and Expected Loss Models used Logistic Regression algorithms that provided the probability of default F1 score = 82%.

• Build Risk Strategy, it refers to a set of rules that are applied to evaluate clients, ensuring that they meet certain criteria before being granted a loan.

• Developed the credit pipeline for products such as microloan, auto credit, educational credit, consumer credit. Average approval rate of loan was 37.5 %, NPL showed 1.3%, within two years, more than 2 trillion sums (180 million USD) loans were issued.

Specialist in Research department — Aug 2018 – Feb 2019

• Multispectral UAV sensors research and engineering of predictive Deep Learning models to forecast with 93% accuracy a dataset with 850,000 measurements and 440 water quality attributes ( NDVI, NDRE, CIR, salinity, chlorophyll, etc.)

• Build a machine learning algorithm, which helps to detect objects and count them. With the help of data processing from UAV, we created an orthophoto of the area that showed the state of objects in this ar, ea.