Olesia

OLESIA TERLETSKA
[open contact info](look above in the "contact info" section)
[open contact info](look above in the "contact info" section) ⋄ LinkedIn ⋄ GitHub

OBJECTIVE
AI Automation Engineer. BSc in AI (CS), ongoing MSc in AI Systems. Saving time and money with AI.

STACK

AI Agents and LLMs RAG Systems, Multi-Agent Workflows, Token Optimization, Stable Diffusion
Workflow Automation n8n, Zapier, Webhooks, REST API Integration, JSON Parsing
Programming languages Python
Databases MySQL, BigQuery, Vector Databases (Pinecone)
Infrastructure and Deployment Google Cloud (SA, OAuth), Railway
Data Science Deep Learning (Transformer), CNN, transfer learning, fine tuning (LoRA)
Version Control Git, GitHub
Agents Claude Code, Hermes
Languages Ukrainian (native), English (C1)

EXPERIENCE
AI Automation Specialist Apr 2026 – till now, 7m
MayaB LLC (full-time)
• Tuned I2I prompts, achieving 89–97% successful generation of product images, saving approx. $1,025 per month
• Optimised the data processing pipeline; resulted in 5.6h of human manual work covered by automation
• Researched, monitored and tested image-to-image generation models, inc. OW and planned the self-hosting
with the LoRA application
• Developed the LLM Assessment pipeline (LLM-as-Judge) assessing the generated images in order to mimic the
human annotator logic and speed up human decision making
AI Developer Oct 2025 – till now, 1y
Tinkogroup (part-time)
• Developed automated data validation system (Python, Railway), reducing manual review time by 16 hours/month
• Developed vector RAG-chatbot for internal use with automatic data collection and user feedback (Pinecone,
Railway, context caching, FSL) with average user rating during 1st month after release - 4.88/5.
• Built an automated image quality control workflow achieving 92% standard compliance rate
• Implemented LLM-as-Judge evaluation pipeline with custom criteria selection; F1 score of 0.80 on assessment
quality
• Introduced Hermes agent to the management team: introduction, MCPs
• Local LLM research: Ollama and Anything LLM for the local RAG support
IT and AI Specialist Dec 2024 – Oct 2025, 10m
Insightvid (project-based)
• Supervised and developed features of the AI-empowered loyalty system platform (the domain of service) for the
LT market, communicated with stakeholders on the requirements and improvements
• Developed the promo website using Web10, Mailchimp, and MC4WP WPForms (WordPress)
• Transformed raw data files into database-compliant formats, streamlining the data migration process and mini-
mizing client downtime (Excel, Google Sheets)

Junior Analyst Nov 2023 – Oct 2024, 1y
B1V (full-time)
• Conducted 86 market analyses on different EU markets, resulting in adding 16 new products to the company’s
assortment; conducted approx. 100 competitor analyses
• Developed automated keyword collection pipeline (PPC campaigns, Python, MySQL), gathering data on 100
products; reduced the process from 16h to 0.5h (96.9% process acceleration)
• Assisted in monthly supply and financial reports (using Excel and Google Sheets; data merging and cleansing;
pivots, xlookup, vlookup, conditional formatting, filters, sparklines)
Intern Jul – Oct 2023, 3m
Reykjavı́k University
• Developed an automated solution for data collection via API from the LMS, which saved 85% of the time required
to complete; performed feature engineering (developed 3 out of 8 features) as a part of academic research

EDUCATION
Master’s degree in AI Systems, Kharkiv National University of Radio Electronics (ongoing)
Bachelor’s degree in AI, Kharkiv National University of Radio Electronics
Diploma topic: ”Transfer learning techniques for the binary classification of cartographical images”

CERTIFICATES
Evaluating AI Agents, DeepLearning.AI
AI Geometry: Understanding How Shape Impacts AI Learning, Udemy
Reinforcement Learning from Human Feedback (RLHF), DataCamp
Data Mining Without Coding: Using RapidMiner in the Context of Education, Tallinn University
Infographics, Projector Institute

ACTIVITIES
Team Lead Feb – Apr 2026
SKELAR x mono AI Competition
• Led the team of 3 people in the competition, assigning the tasks, collecting the feedback and shaping the report
Researcher May 2026
XXXIV International Scientific and Practical Conference. Information Technologies: Science, Engineering, Technol-
ogy, Education.
• Compared local vs cloud (Groq) Llama 3.1 8B on 1,000 QA and EN↔UA translation tasks (TTFT, TPS, latency,
LLM-as-judge): cloud was ∼40–60× faster, while local reached ∼95% of its quality (3.7 vs 3.9/5)