AI developer and consultant based in Budapest. I build custom AI agents and business automation for Hungarian SMEs and international clients.
TL;DR: 3.5+ years building custom AI systems. Specialties: AI agents (OpenAI Function Calling), accounting and customer-service automation, n8n workflows, RPA. Largest measured impact: ~80% reduction in admin load at a 5-store Hungarian SME (Novetex). Working languages: Hungarian (native) and English.
What I do
Custom AI agents — built on OpenAI GPT-4o / o1, Anthropic Claude, Google Gemini with Function Calling + RAG.
Business process automation — n8n workflow design, Python backends (FastAPI/Flask), RPA (Playwright, Selenium).
Accounting automation — GPT-4 Vision invoice OCR, ledger coding, bank ETL, SAP / SQL Ledger integrations.
Customer-service AI — multi-language email assistants, chatbots, AI voice agents (Twilio + OpenAI Realtime API).
AI training & consulting — developer and executive workshops on prompt engineering, AI agents, use-case discovery.
Experience & references
Novetex (5-store retail network) — full claims and admin digitalization with Python + OpenAI Assistant API. Result: ~80% reduction in admin load.
FinTech automation — n8n parallel workflow orchestrating 4 target systems (Excel API, SQL Ledger, MS Teams, email) with GPT-4 Vision invoice processing. Result: ~80% reduction in manual data entry.
Real-time AI voice assistant — Twilio Voice + OpenAI Realtime API, async FastAPI WebSocket backend, direct audio-to-audio.
CLB Insurance Brokers — corporate AI developer training (ChatGPT Prompt Engineering for Developers, Multi-Stage System Building).
Aura Interiors — premium portfolio website design and implementation.
Tech stack
AI / LLM: OpenAI API (GPT-4o, o1, Vision, Realtime), Anthropic Claude, Google Gemini, Llama, Mistral, LangChain, prompt engineering.
Frontend: Modern web (HTML5/CSS3, TypeScript), React, SwiftUI (native macOS).
DevOps: Git, Docker basics, Linux, ngrok.
How I work
Practical, not academic: I build working systems that make money, not theoretical models. Iterative and agile — a working first version in weeks, then tuning based on real usage. Every critical decision goes through human-in-the-loop checkpoints and audit logs.