Марян

MARYAN DIDUKH
AI Engineer — GenAI, LLM Applications, Agentic Systems & RAG
[open contact info](look above in the "contact info" section) | [open contact info](look above in the "contact info" section) (PL) · [open contact info](look above in the "contact info" section) (UA) | github.com/Marian003
English C1 · Polish B2 · Ukrainian (native) · Available in Ukraine / Poland · Remote or Warsaw / Lviv-based

🏆 UA SKILLS "AI-SKILLS" Tournament 2026 — 2nd place of 212 participants (prize winner, $5,000 fund). Judges' verdict: "surgical,
precise application of LLMs and flawless flow logic in the interface" — 29/30 for live functionality.

SUMMARY
AI engineer with 2 years of commercial experience who builds software with AI, about AI, and for AI-driven workflows. Recently
ranked 2nd of 212 in a national AI engineering tournament for precision LLM application. My daily toolkit is agentic: Claude Code with
custom slash commands and MCP servers, multi-model orchestration across Claude, GPT-5, Gemini, rapid-build platforms (Base44,
Replit, Lovable), and local Ollama models. I design prompts and RAG pipelines, build guardrails against hallucination and data leakage,
and ship LLM features end to end — discovery, implementation, evals, tests, and deployment. Strong analytical side: log-driven root-
cause analysis, edge-case hunting, and technical documentation as part of knowledge management.

TECHNICAL SKILLS
AI / GenAI: Anthropic Claude API, OpenAI GPT-5, Google Gemini, Ollama (local LLMs); prompt engineering & evaluation, RAG
(chunking, retrieval, eval loops), multi-agent orchestration, guardrails (hallucination control, token/context management)
Agentic tooling: Claude Code (custom slash commands, custom tools, subagents, agentic workflows), MCP servers (building &
integrating), AI-assisted SDLC with ChatGPT / Gemini / Claude, rapid prototyping (Replit, Lovable, Base44)
Languages: Python, TypeScript, JavaScript, C#, SQL, PHP, Bash
Python stack: FastAPI, pandas, numpy, httpx, asyncio, pytest, scikit-learn (basics), CLI tooling (argparse, click)
Web: React, Next.js 15, Vue.js, TailwindCSS, REST APIs, Tiptap/Monaco editors
Data & DevOps: PostgreSQL, Oracle, Redis, Prisma ORM; Git/GitHub, GitHub Actions CI/CD, Docker, Turborepo, pnpm, Postman, Jira,
Confluence; pandas, plotly, Power BI (basic)

AI COMPETITIONS
UA SKILLS — "AI-SKILLS" National Tournament (prompt engineering / AI-first development) May–Jul 2026
• Placed 2nd out of 212 participants (18 finalists) in a competition for senior-level developers, judged on prompt engineering depth
and architecturally sound, fully functional AI solutions.
• Official judges' feedback: "surgical, precise application of LLMs and flawless flow logic in the interface" — scored 29/30 for live
functionality; Node.js / TypeScript / React / Tailwind stack.

EXPERIENCE
AI-First Frontend Developer — Devocean (remote, healthcare & logistics clients) Oct 2025 – Present
• Deliver client features through an agentic Claude-based workflow: Claude Code with custom slash commands and MCP servers
wired to the repos, ticketing, and docs — planning, generating, and reviewing code with AI in the loop at every step.
• EmCell (regenerative medicine): built an internal web app for stem-cell passport management and patient records — responsive
forms, search views, data-entry flows, and dashboards for cell tracking and treatment history used daily by medical staff.
• Gralsen (logistics): shipped shipment/dispatch workflow features, order tracking views, and internal reporting screens; co-
developed a custom, locally trained AI system for parsing 100+ different document types used in logistics operations.
• Work directly with clinicians and operations teams to translate domain requirements into UI that doesn't get in their way.
Full Stack / AI Developer — Vovos / Media Projects (35mm.online) May 2024 – Sep 2025
• Delivered Vue.js features (content views, filtering, search, accessibility, responsive fixes) across multiple release cycles using an AI-
assisted workflow built on ChatGPT and Gemini — from spec analysis and code generation to test scenarios and debugging.
• Worked on C#/.NET + Python backend services and Umbraco CMS templating; ran the Umbraco v8 → v13 migration, using LLMs
to map content structures and draft the deployment runbooks the team still uses.
• Owned support: ticket triage and log-based root-cause analysis with LLM-assisted log summarization; recurring incidents dropped
~30% after I documented the top failure modes.
• Built and validated REST API integrations in Postman, wrote integration test scenarios, coordinated UAT; prototyped UI concepts
with Replit, Lovable, and Base44 before committing to full development.

KEY SOLO PROJECTS — BUILT END TO END WITH AGENTIC AI WORKFLOWS
Mindlink — Multiplayer Human + AI Workspace · Next.js 15, Hono, Claude, OpenAI, Yjs, PostgreSQL, Redis 2025
• Collaborative workspace where humans and AI agents share one document in real time — agents reference each other, debate,
and vote on decisions. Not a chatbot, not a copilot.
• Designed 6 specialized agents (architect, code reviewer, PM, user researcher, security auditor, creative ideator) routed across
Claude and GPT-4o, each with a distinct system prompt, personality sliders, and behavior rules.
 • Real-time CRDT sync (Yjs + y-websocket); document/code/whiteboard modes via Tiptap, Monaco, Excalidraw; Redis + BullMQ
queueing; Turborepo monorepo. Developed with Claude Code using custom slash commands and MCP servers for repo,
database, and deployment automation.
ReviewBot AI — AI-Powered Code Review Platform · Python/FastAPI, React/TypeScript, Claude API 2025
• Dual-analysis review: rule-based static scanning + live AI review via the Claude API; prompt templates and guardrails against
hallucination, off-topic output, and code/data leakage.
• Delivered end to end with a Claude-driven agentic workflow: discovery, requirements, implementation, tests, landing page,
deploy.
Elvish Language Creator — LLM / RAG Research (Diploma) · Python, Ollama, local LLMs 2024
• RAG pipeline for a constructed-language use case: chunking, context injection, prompt templates, iterative eval loop; reliability
and format-compliance testing against local LLMs, with prompts refined from observed failure modes.
GitPulse CLI — Developer Productivity Tool · Python, httpx, pytest, GitHub Actions 2024
• Async CLI dashboard for GitHub analytics (streaks, language breakdown, repo metrics); pytest + respx suite, CI/CD via GitHub
Actions; built with an AI pair-programming workflow (ChatGPT/Gemini).

EDUCATION & CERTIFICATIONS
Polish-Japanese Academy of Information Technology (PJAIT), Warsaw — B.Eng., Computer Science, Feb 2026 · Major: Intelligent
Data Processing Systems · Final grade: dobry plus · PRK level 6
Ternopil Ivan Puluj National Technical University — Computer Science (graduated) · Lviv IT School (LITS) — Web Technologies (2020)
Anthropic — Claude Code courses (agentic coding workflows, tool use, prompting) · DataCamp — Associate Data Analyst in SQL
Focus: GenAI delivery · agentic development · prompt engineering · clean documentation