Profile
I build AI agents that run real operations. AI Engineer with 6 years of experience designing, deploying and operating automation and Generative AI systems in production: LLM agents, RAG, multi-agent orchestration and workflow automation, owned end to end from process analysis to evals, guardrails and monitoring. A Business Analyst background means every project starts from a KPI and ends with a measured result. Hands-on problem solver and team player; AI should absorb the routine so people can focus on work that needs judgment.
Experience
AI Automation Engineer · Team Lead since Jan 2026
- Problem
- ~4,000 tickets a day handled by hand; under overload a first reply took up to 18 min.
- Built
- A chat widget that answers in the customer's language, checks logged-in accounts, solves requests and escalates complex cases to a human in Zendesk with full context.
- Stack
- n8n · LangChain / LangGraph · GPT-4o mini · Pinecone · MongoDB · Redis · Langfuse
- Result
- 70%tickets solved by AI, 24/718 min → 3 sfirst reply2.5 → 3.8CSAT0.6 → 0.9RAG score
- Problem
- Support, sales and management used separate tools and learned about problems late.
- Built
- One CRM where every team sees its customers, gets an alert the moment something happens and uses an AI assistant that routes requests and answers from the knowledge base; QA per milestone, Zendesk as benchmark.
- Stack
- TypeScript · Kubernetes · Kafka (with ML team) · AWS · Tableau · Voiso · Telegram
- Result
- 1,000+users10–20kmessages a month0pending alertsMVPdelivered, in final QA
- Next
- On-premise LLM (Qwen 3, Qdrant) with ML and DevOps: data stays in-house, lower cost and latency, sensitive-data filter before external LLM calls, fine-tuning in preparation.
- Problem
- Dozens of brand websites needed every month, with no content team to build them.
- Built
- A pipeline of agents that researches a topic or example site, writes the content, generates images, applies SEO and publishes the site on the client's servers.
- Stack
- Hermes agents · MCP · Claude Code · Codex · Perplexity · GPT-6 Astra
- Result
- ~100sites a month5–10 minper site~€5cost per site
- Problem
- Routine HR, Finance and Legal questions from employees; no self-service on the website.
- Built
- One assistant per department in Teams answering from company documents; a website assistant answering from the knowledge base and capturing leads by email.
- Stack
- Copilot Studio · Azure OpenAI · Databricks · SharePoint
- Result
- 3departments covered 24/7FAQ + leadshandled on the website
AI Automation Agent Developer
- Problem
- Leads arrived by email and managers replied late, losing buyers.
- Built
- A website assistant that presents properties, collects the buyer's contacts, alerts the manager on Telegram, sends a follow-up and books a meeting in one click.
- Stack
- OpenAI · Make · Airtable · Voiceflow · Telegram · Google Calendar
- Result
- +15%leads+15%client revenue5managers onboarded
- Problem
- Clients needed to answer customers at any hour without hiring staff.
- Built
- Chatbots on WhatsApp, Instagram and Telegram that qualify each contact into the CRM; voice agents for HR, sales and outbound calls.
- Stack
- OpenAI · Anthropic · n8n · Apps Script · Vapi · Retell · Twilio · ElevenLabs
- Result
- 1,000+conversations a monthcold → hotpipeline automated
- Problem
- Clients needed websites fast; agency processes stopped scaling past 200 people.
- Built
- 20+ client websites designed, built and deployed end to end; all agency processes rebuilt in ClickUp (HR onboarding, analytics, scheduling).
- Stack
- ClickUp · Google Workspace · Next.js · Astro
- Result
- 20+websites delivered50 → 200people supported
Business Analyst
- Result
- −25%idle stock€2–3M/yrsupplier invoicing controlled4analysts trained
Certifications Hugging Face AI Agents · DeepLearning.AI Evaluating AI Agents & Prompt Engineering · Anthropic Academy Claude API · LangChain Academy LangGraph · Make (Web3Unity) · n8n (Mask)
Languages Ukrainian native · Italian C2 · English C1 · Russian C1 · Spanish B1