Maksym

MAX BADYULA
AI & Automation Engineer | AI Agents · LLM Integrations · Python · n8n
Mykolaiv, Ukraine · [відкрити контакти](див. вище в блоці «контактна інформація») · [відкрити контакти](див. вище в блоці «контактна інформація»)
[відкрити контакти](див. вище в блоці «контактна інформація») · github.com/VETERAN1945

PROFESSIONAL SUMMARY

AI & Automation Engineer building LLM-powered agents and internal automations for real business tasks. Works on both levels: fast MVP delivery in
n8n, and agent internals written in pure Python — tool-calling loop, session memory, RAG retrieval and error handling implemented by hand rather
than taken from a framework. Backed by a Software Engineering degree and a working foundation in REST APIs, SQL and end-to-end testing,
applied to make automations reliable, debuggable and documented.

CORE COMPETENCIES

AI & Agents: AI agents (tool-calling, multi-step workflows), RAG (retrieval-augmented generation), text embeddings & vector similarity search,
system-prompt and tool-description design, LLM output parsing, structuring and validation, MVP / PoC prototyping
LLM APIs & Tooling: OpenAI API, Claude API, Groq API; Cursor, Claude Code (AI-assisted development)
Python: FastAPI, uvicorn, Pydantic (request validation), sentence-transformers, NumPy, python-dotenv, requests, JSON serialization, virtual
environments and packages
Automation & Low-Code: n8n (AI Agent node, visual workflows, webhook triggers, HTTP Request nodes, multi-branch logic), Make.com, Zapier
APIs & Protocols: REST API, Webhooks, JSON, HTTP/HTTPS
Databases: SQL, PostgreSQL, MySQL (joins, subqueries, data extraction)
Other languages: JavaScript (custom functions in n8n), basic TypeScript, Bash
Testing: Playwright (E2E), Postman (API test automation, JS assertions, Collection Runner)
Infrastructure: Linux/Ubuntu (CLI, SSH, filesystem), Git/GitHub
Languages: English (B1+, technical documentation), Ukrainian (native), Russian (native)

AI AGENT PROJECTS

RAG Support Agent, built from scratch — Python · FastAPI · Groq · sentence-transformers · NumPy
Personal engineering project: rebuilt an existing n8n support agent in pure Python, without agent frameworks, to own every layer of the mechanics.
Built the agent as an HTTP service (FastAPI + uvicorn): request validation with Pydantic, per-user session storage, model calls and response
handling.
Implemented the tool-calling cycle manually — model's tool request → execution in code → role: "tool" message → second completion — instead
of delegating it to a framework.
Added RAG over an internal knowledge base: sentence-transformers embeddings built once at server start, retrieval by vector distance in
NumPy. The document is selected deterministically by code; the model only interprets and phrases it.
Diagnosed and fixed a silent retrieval failure: an English-only embedding model produced a 0.005 distance gap on non-English documents, i.e.
selection by noise. Switching to a multilingual model raised the gap to 0.63–1.24 and corrected document choice on all test questions.
Improved retrieval by searching with the user's original message instead of the model-generated query string, removing a non-deterministic step
between question and document.
Made the service resilient to non-deterministic LLM behaviour (malformed tool calls, expired API key): try/except around both model calls plus
rollback of session history to a pre-request snapshot, so a failed request returns a valid response and leaves no orphaned messages in the
session. Verified against 4 real failures in a single test run.
Documented known limits and next iterations: no relevance cut-off in retrieval, no system-prompt grounding, no retry on malformed tool calls,
sessions lost on restart.
Repo: github.com/VETERAN1945/python-ai-agent-support

Affiliate Partner Support Agent — n8n · LLM · tool-calling
Built an autonomous AI agent that handles incoming partner (webmaster) messages and decides, per message, which tool to call — payout
database, tracking-link checker, or offers list — using LLM tool-calling instead of a fixed flow.
Designed the agent from scratch: defined its tools and their descriptions, wrote the system prompt, and added conversation memory for multi-
turn context.
Integrated a chat model (Groq/OpenAI) with a data layer built on Google Sheets and HTTP Request nodes.
Repo: github.com/VETERAN1945/ai-agent-affiliate-support

Campaign Diagnostics Agent — n8n · LLM · reasoning loop
Built an AI agent that investigates drops in campaign performance through a reasoning loop: it pulls campaign stats, then chooses what to check
next (landing-page availability, payout-history changes) based on intermediate results, and reports the most likely root cause.
Implemented multi-tool orchestration where the investigation path is selected by the model at runtime rather than hard-coded — agentic
behaviour vs. a linear pipeline.
Repo: github.com/VETERAN1945/ai-agent-campaign-diagnostics

AUTOMATION & LLM WORKFLOWS

AI-Driven Workflow Automation — n8n & LLM integrations
Built multi-step self-hosted n8n workflows using Webhooks and HTTP Request nodes to parse raw incoming data.
Integrated Claude and OpenAI APIs to extract key parameters and map them into clean, validated JSON structures.
Designed error-handling branches and Telegram/Slack alert triggers for reliability and instant incident reporting; synced real-time data between
webhooks and target systems (Google Sheets, SQL).

SEO Automation, Web Scraping & API Monitoring — n8n
Built scheduled n8n pipelines for SEO data collection, news scraping, and automated REST API health monitoring with alerting and structured
output.
Repos: github.com/VETERAN1945?tab=repositories

TESTING & QA (SUPPORTING)

Web Automation Framework (Playwright) — developed a scalable E2E testing framework with dynamic locators and robust assertions to
validate critical user paths and UI elements.
API Testing Automation (Postman) — created automated test suites for REST endpoints (GET, POST, PUT, DELETE) with pre-request scripts,
 environment variables and JavaScript assertions; ran batch executions via Collection Runner.

CERTIFICATIONS

Python Fundamentals — SoftServe Academy (December 2024)
Database Fundamentals — SoftServe Academy (October 2024)
Linux & Networks Fundamentals — SoftServe Academy (June 2024)

EDUCATION

Junior Specialist in Software Engineering — Mykolaiv Building Professional College of KNUCA (2016–2020)
Relevant coursework: SDLC, SQL & database design, system analysis, algorithms, software testing.

WORK EXPERIENCE

Equipment Operator — Pivden Bud (2018–2024)
Executed precise technical checklists and isolation protocols, building a rigorous root-cause and quality mindset later applied to debugging and
testing.
Diagnosed complex equipment malfunctions through systematic troubleshooting; maintained detailed incident logs and reports — a baseline for
structured documentation and runbooks.
Transitioned into IT through dedicated self-study while in full-time employment.