I'm a QA Automation Engineer and researcher with expertise in AI-driven test systems, large language models (LLMs), vector search, and cloud infrastructure on AWS. Currently working as QE Engineer at Resolver Inc. (Kroll) , where I lead software quality engineering and AI-assisted automation research.
I am a co-author of a Springer Nature book chapter on data-centric AI for psychological support. My research interests span intelligent test automation, LLM integration in enterprise software, and cloud-native QA pipelines. I hold AWS certifications and I am a member of the IEEE Foundation.
Experience
Department of Development / Software Quality Assurance. Leading QA automation strategy and AI-assisted testing research. Internal research project on AI-Assisted QA Automation (2025).
Projects
A TypeScript tool that scans public GitHub repositories for exposed AWS credentials and Bedrock-related secrets, validates discovered key pairs against live AWS APIs, and generates structured JSON reports.
Built a fully reproducible pipeline that scrapes official U.S. immigration sources (USCIS, 8 CFR, BIA) and generates a 17,058-question Q&A dataset using AWS Bedrock Claude. Fine-tuned Llama 3.2 3B via SageMaker JumpStart with LoRA (r=32), producing a model that outperforms the Llama 3 8B zero-shot baseline by +27% mean score and delivers 4× more fully-correct answers — at one-third the parameter count.
17,058 source-grounded Q&A pairs from the immigration law dataset. Answer retrieval
uses keyword-overlap scoring with stopword filtering, returning top matches with clickable
source URLs and authority levels. Runs entirely on CPU (zero LLM inference) — no API keys,
no model hosting, zero cost.
Large-scale Q&A dataset covering U.S. immigration law across 13 subdomains (family-based, naturalization, asylum, removal, and more), built entirely from official government sources. Each record carries source_url, source_span, and authority_level for citation-aware RAG and fine-tuning. Includes 10,056 canonical source documents in the corpus config.
View on HuggingFace · DOI: 10.57967/hf/8824
Llama 3.2 3B Instruct fine-tuned with LoRA (r=32, all attention projections) on 16,065 immigration law Q&A pairs via AWS SageMaker JumpStart. LoRA weights are merged into the base model — load with standard AutoModelForCausalLM, no adapter setup required. Evaluated with Claude Sonnet 4.6 as LLM-as-Judge on 101 held-out questions.
A zero-config, plug-and-play end-to-end testing framework featuring 7 CLI agents, AWS Bedrock LLM-as-Judge scoring, semantic assertions, hallucination detection, AI locator fallback, MCP server integration, security fuzzing (OWASP ZAP), mutation testing (Stryker), multi-role auth matrix, i18n, accessibility, and CI/CD pipelines with Jira sync and Slack reporting.
Batch-converts SwitchBot camera SD card recordings (.media + .info files) into standard MP4 videos with audio — drag, drop, done.
Education
Department of Economics and Finance
Vinnytsia College of Trade and Economics of Kyiv National University of Trade and Economics · vtec.vn.ua
Publications
Peer Reviews
Verified peer review · Publons ↗ ·
Verified peer review · Publons ↗ ·
Verified peer review · Publons ↗
Certifications & Distinctions
Memberships
Chapter contribution to Lecture Notes on Data Engineering and Communications Technologies. springer.com ↗