Личные данные скрыты

Roman Nalapko
AI Engineer · Automation Specialist · Python Developer

[открыть контакты](см. выше в блоке «контактная информация») · @BEATZIRO · Romania / Ukraine (Remote) · LinkedIn · English B1

ABOUT ME

I build automation systems and AI tools that cut real manual work — bots, pipelines, scrapers, LLM integrations. Started
with Python and Telegram bots back in 2022, and over time moved into AI tooling: local LLMs, OpenAI API, n8n, Make.com.
Most of my projects came from actual problems — a friend running a crypto course needed 24/7 student support without
hiring staff, an SMM team needed content published across platforms without touching a computer. I figure out what’s
eating the most time, then automate it. Currently working with: Ollama (local LLM deployments), OpenAI / Claude API,
n8n, Make.com, Whisper, Python async stacks. Remote only. Ukraine / Romania.

TECHNICAL SKILLS

AI & LLMs OpenAI API (GPT-4/4o), Claude API (Anthropic), Ollama (local LLM on server), Whisper / faster-
whisper, TTS, prompt engineering, system prompt design

Automation Make.com, n8n, Zapier, Telegram Bot API, SmartSender, webhook pipelines, multi-platform content
distribution

AI Dev Tools Cursor, Windsurf, AI-IDE / vibe coding, agent-assisted development

Python asyncio, aiohttp, requests, BeautifulSoup, lxml, pandas, aiogram

Scraping Playwright-stealth, proxy/session management, bot-detection bypass, captcha handling

Desktop / GUI PyQt5/6, threading, MVC architecture, PyInstaller

Databases PostgreSQL, MySQL, SQLAlchemy

DevOps Docker, Docker Compose, Linux, ETL pipelines, multiprocessing

Other API integration, XLSX export, deduplication, retry logic, structured logging

PROJECTS

AI FAQ Bot — Crypto Course Assistant 2024
Private Telegram bot with a local LLM serving as a 24/7 student support assistant Tech: Ollama (local LLM on server),
Telegram Bot API, Python, aiogram, custom knowledge base pipeline
• Set up Ollama on a private server so the model runs locally — no data sent to OpenAI, full privacy
• Fed the entire crypto course content into the knowledge base and tuned the system prompt so the bot
only answers from that material
• Sat down with the client, went through his workflow, and figured out that FAQ replies were eating most of
his day — that’s what we automated
• Bot ran in Telegram 24/7, students got instant answers, client stopped answering the same questions
manually
Result: Response time: hours → instant; support load cut by ∼70%; ran in production

AI Content Distribution Pipeline — n8n + ChatGPT 2024
Automated social media content generation and multi-platform publishing Tech: n8n, OpenAI API (ChatGPT), Instagram
API, TikTok API, webhook triggers, JSON processing
• Built the flow in n8n: webhook fires, ChatGPT generates the post, output gets formatted differently for
Instagram vs TikTok, then posted via API
• Wrote separate prompts per platform — captions for Insta are different from TikTok hooks
• Added retry logic so if the API rate-limits, it just tries again instead of failing silently
Result: Publishing: 20–30 min manual work → zero; tested end-to-end, works as a full pipeline

Speech-to-Text AI Telegram Bot (Open Source) 2025
Voice transcription bot using a local Whisper model with SRT subtitle generation Tech: aiogram 3.7+, faster-whisper,
FFmpeg, asyncio, ThreadPoolExecutor, Python 3.9+
• Used faster-whisper instead of the standard one — same quality, runs about 4x faster, works fully offline
• Converts audio to 16 kHz WAV first, then transcribes in parallel threads so long files don’t block the bot
• Detects language automatically, can output .srt subtitles if needed
Result: 3–5 min audio done in 1–2 sec; personal note-taking got way faster; open source on GitHub

Instagram Auto-Poster — Telegram + Make.com + Dropbox 2025
No-code AI content pipeline for SMM teams Tech: Make.com (webhooks, JSON), Telegram Bot API, Dropbox API, GPT-
assisted templating
• SMM team was spending 5–10 minutes per post just on file prep and uploading — built a Telegram bot
where they just send the content and it handles the rest
• Make.com scenario picks it up, structures the metadata, drops files into Dropbox automatically
• No technical knowledge needed on their end — just send a message
Result: Post prep: 5–10 min → 10–15 sec; team saves 20+ hours/month

Instagram Data Collector & Scraper 2025
Desktop ETL tool for authorised Instagram content collection with AI-ready output Tech: Playwright-stealth, aiohttp,
BeautifulSoup, lxml, multiprocessing, XLSX export, ETL
• Built the whole anti-detection stack myself — stealth headers, rotating proxies, random delays, session
handling
• Runs multiple profiles in parallel via ProcessPoolExecutor, deduplicates, retries on failures
• Outputs clean XLSX — analysts just open it and work, no cleanup needed
Result: 300+ profiles/hour; cut analyst manual work by 90%+

Telegram Automation Toolkit 2025
Desktop app for audience growth — parsing, inviting, CRM-ready export Tech: PyQt6, Telethon, asyncio, QThread,
SOCKS5 proxy, .session file management
• Parses Telegram groups async via Telethon, has built-in delays and anti-spam logic so accounts don’t get
flagged
• If it gets interrupted mid-run it picks up where it stopped — doesn’t start over
• PyQt6 interface so non-devs can use it without touching the terminal; SOCKS5 proxy support built in
Result: 1000+ users/hour; saves 40+ hours/month for growth teams

LuxWash AI Assistant Bot 2024
Automated customer communication for a self-service car wash Tech: SmartSender, Telegram Bot API, trigger/flow
automation
• Car wash was getting flooded with calls asking ”is there a queue?” — built a bot that shows the queue,
takes feedback, sends promo info
• Customer just taps through a menu, shares location, gets all the info — no one picks up the phone
Result: Calls dropped 50–65%; repeat customers up ∼20%; 3x more feedback collected

ChipDale Telegram Bot — Startup MVP 2022
End-to-end task request automation for an on-demand services startup Tech: Python, pyTelegramBotAPI, OOP, multi-
step form logic
• Startup needed an MVP fast — built a bot that walks users through a multi-step form (cleaning, moving,
repairs, delivery) and forwards the structured request straight to the operator
• Whole request flow in Telegram, no app install, no calls
Result: Request time: 5–7 min → 40 sec; 100 tasks/day; operator load dropped 60%

skliarova.com — Commercial Website 2023
Client site with UX-first design, SEO and brand presentation Tech: Wix, SEO tools, responsive layout, analytics integra-
tion
• Built together with a designer — I handled structure, content hierarchy, mobile layout, and basic SEO
• Set up analytics and conversion tracking from day one
Result: Conversion up 12–18%; organic traffic up 15–30% after launch

EDUCATION

Navigation — Maritime University 3rd year, current
Everything in AI and automation I picked up myself — through real projects, breaking things, and fixing them.

WHAT I BUILD

Local LLM setups (Ollama) · Telegram bots with AI backends · n8n / Make.com automation pipelines · Scrapers and ETL
tools · Voice transcription (Whisper) · Desktop apps with PyQt