01 / 06
mono-api-agent
RAG agent over monobank's official OpenAPI specification: chunking per method, hybrid retrieval and a deterministic hallucination check instead of an LLM judge.
View projectOpen to junior roles in data and AI engineering · remote or in Bavaria
Oleksandr Shevchenko
Data & pipelines · AI agents · AI-assisted
Data preparation · RAG & retrieval · automation · Python
For about three years I have worked daily with data that is dirty, constantly changing and still has to be processed. That turned into a working method: measure before claiming anything, and rebuild when the measurement comes back bad. My core project is mono-api-agent — a RAG agent over an official OpenAPI specification, built in one day and rewritten twice that same day because the numbers demanded it. I have not trained models in production — that is the area I want to learn.
Selected projects
Practical projects. Clear focus.
The flagship: a public Playwright test suite that checks this very website in CI — 56 checks across desktop and mobile. Alongside it, four product demos from more than 20 personal projects; every demo can be tried directly in the browser.
View the test suite on GitHub02Portfolio E2E test suiteMy own Playwright + TypeScript E2E suite testing this website in CI: UI, API and accessibility — desktop and mobile.View03AI Application HelperAnalyses job ads and generates personalised cover letters with an editable export.View04SmartChatModern AI chat with streaming, history, markdown and code highlighting.View05DevDashCustomisable developer dashboard with widgets, Pomodoro and GitHub activity.View06Rent Price TrackerCompares rent costs across German cities with trends and an income calculator.View
Practice & growth
From self-study to working systems.
Three years of daily automation practice — more than 20 projects, over 100 scripts and, since 2026, a public e2e suite in CI. With an honest view of strengths and knowledge boundaries.
2018 — 2022
Bachelor in Economic Cybernetics — ZAB-recognised
Degree from the State University of Trade and Economics (Kyiv) — mathematical modelling and information systems, recognised by the ZAB as equivalent to a German bachelor (03/2025). Followed by 1st place among ~6,000 participants in the RS School / EPAM JS/FE Pre-School course.
Foundation: degree · EPAM RS School top 12023 — today
Own software & automation projects — AI-assisted
Built and operated my own projects daily, consistently on data that was dirty and constantly shifting. Verification, DevTools and my own traffic analysis — 20+ projects, 100+ scripts and workflows, kept reliably in operation over months.
Focus: messy data · verification · reliability2026
Focus: data & AI engineering
Built and published mono-api-agent: a RAG agent over monobank's official OpenAPI specification, created in one day and rewritten twice that day because the measurements demanded it — chunking per method, embeddings over the method's purpose, a deterministic hallucination check (60 s → 2.8 s, five model calls → two). I am looking for a junior role where I can develop this further inside a team.
Goal: data quality, retrieval, applied AI
About me
I orchestrate AI. I take responsibility for the result.
After my degree in Economic Cybernetics (ZAB-recognised bachelor, meaning mathematical modelling and information systems) I finished 1st among roughly 6,000 participants in the RS School / EPAM JS/FE Pre-School course. It is the only number about me that can be verified in a second — and mostly it says one thing: I learn fast when I have something concrete in front of me.
Since then I have spent about three years working daily on my own software and automation projects — AI-assisted: I define the requirements, drive coding agents, verify the result and debug through DevTools and network traffic analysis. The input data was never clean and rarely stable. 20+ projects and 100+ automated workflows, kept reliably in operation over months. When an unstable workflow costs money directly, verification becomes a habit.
Today I am pointing that combination at data and AI engineering. My core project is mono-api-agent: a RAG agent over monobank's official OpenAPI specification, built in a single day. Chunking per API method instead of by character count, embeddings computed over the method's purpose instead of the full chunk, a deterministic hallucination check instead of an LLM judge. That same day I rebuilt twice what already worked, because the measurements showed it worked badly. I have not trained models in production — that is the area I want to learn.
Data & pipelines
Data preparation & cleaning · Spotting broken and incomplete records · Format conversion & validation · Python for automation and scripting · Running pipelines under shifting inputs
AI, agents & RAG
RAG: chunking strategies, embeddings, vector search · Hybrid retrieval (lexical signal over vectors) · Deterministic checking instead of an LLM judge · Multi-agent workflows (LangGraph) · Prompt & context engineering
Automation & testing
Playwright (public E2E suite) · Browser automation · DevTools & network analysis · Git / GitHub Actions (CI) · Failure analysis & recovery
Machine learning
Reading and interpreting metrics · Interpreting training curves · No models of my own trained in production · Current learning focus
Contact
Let’s get to know each other.
I am open to junior roles in data and AI engineering or an initial technical conversation.
oleksandr.o.shevchenko@gmail.com