Герман

Herman Maliarenko
AI Automation Engineer

Sveti Vlas, Bulgaria · remote (EU time) · [відкрити контакти](див. вище в блоці «контактна інформація») · [відкрити контакти](див. вище в блоці «контактна інформація») ·
aerra.io

Automation engineer building AI systems on n8n, Python and JavaScript: agent pipelines, RAG, scrapers,
API integrations. ~2 years of hands-on AI and automation work with real clients. I design, build,
document and hand off production systems.

SELECTED COMMERCIAL WORK

Lead-generation pipeline for the game industry aerra.io, 2026
Scrapers for Steam and expo catalogs with rate-limit backoff and resume-safe runs.
Deterministic ICP scoring, then AI verification agents checking every lead against open sources (site,
Steam, press).
134 verified leads with evidence links went out in the first month, each with an outreach draft, instead
of checking studios by hand.

Indie game studio Germany
AI support agent on a Supabase knowledge base. Junk was filtered at intake and replies were drafted
from the studio's docs, so the team was not writing every answer from scratch. Ticket counts went to
Discord.
Go email-to-ticket service: Gmail threads became tickets in their admin panel, grouped by thread and
matched to a player, instead of support staying in the inbox. 41 unit tests.

Multi-brand content creator US
Six n8n workflows for three creator brands. Each day a scored post, brand-voice copy and an image
landed in Buffer as a draft, so nobody had to make that post by hand.

E-learning client repeat client
n8n scored Upwork jobs and emailed a proposal ready to send, so those proposals were not written
from scratch.
Flask API pulled structured rows out of messy Storyline .docx into Google Sheets, so course text was not
copied by hand.

Game-art studio first client
Content pipeline from source material to assets ready to publish, so that production step was not done
by hand.

OWN PRODUCTS

Job Lens: opening a job turns into a fit score and a proposal draft. A second model checks the draft
before it is shown. Chrome extension, routing across Claude, OpenAI, Gemini and Groq, Telegram
alerts.
Market Angel: a research question is split across agents and the findings stay in one place, instead of
a manual search across tabs.
Cold-outreach system: a lead is enriched, the email is checked by a second model, and follow-ups go
out without writing each one by hand.

Voice Typer: speech becomes text in any app on the machine, without a cloud subscription.
Video dubbing pipeline: training videos with a heavy accent were re-voiced into a clear narration for
a medical-device distributor whose customers are doctors.
Voice assistant on aerra.io: a visitor question on the site gets an answer from the assistant, without
waiting for a person.

STACK

Automation n8n (self-hosted on Hetzner), Make
AI (LLM) Model-agnostic via OpenRouter and direct APIs: Claude, OpenAI, Gemini, GLM, Qwen
and other US / Chinese models; agents, RAG (Supabase pgvector), MCP, multi-model
routing with LLM-as-judge validation
AI (image / Nano Banana, Qwen, Flux, LoRA pipelines; video: Seedance, Kling (via Wavespeed and
video) direct APIs)
Voice AI ElevenLabs (TTS, STT, dubbing, conversational agents), VAPI, Retell AI
Code Python (Flask, scrapers, ETL), JavaScript / Node, TypeScript / Next.js basics
Dev tools CLI-first workflow; Claude Code, Codex, Cursor, Windsurf, Antigravity, VS Code
Data / infra Supabase / Postgres, Docker (working level), Hetzner, Vercel
Tools Apify, Telegram bots, Google Workspace APIs, Buffer

LANGUAGES

Ukrainian, Russian: native. English: intermediate, daily written use.