About

I am a System Architect based in Budapest, Hungary. My engineering work sits at the intersection of two traditionally separate disciplines: power electronics hardware design and deterministic, validated agentic software execution.

My primary focus is bringing AI augmentation and rigorous systems engineering together to build hardware and runtime environments that cannot afford silent failures.

Why this publication exists

Autonomous agent systems and AI-driven workflows are transforming engineering, but they often lack the determinism, safety boundaries, and physical reality checks required for mission-critical applications.

Coming from an electrical engineering and systems background, I approach software, agents, and hardware with a single operating principle: if a process isn't deterministic, bounded, and verifiable against physical constraints, it isn't production-ready.

This blog serves as an engineering log, architectural catalog, and benchmark record covering:

• AI-Augmented Power Electronics Design: Leveraging LLM tooling, closed-loop simulation agents, and ML surrogate models to accelerate power stage optimization—from automated magnetics sizing and thermal/EMI boundary exploration to SPICE-in-the-loop verification and topology selection.
• Untrusted Code Boundaries & Sandboxing: Runtime isolation architectures (e.g., gVisor/runsc, low-overhead microVMs, and container hardening) designed to run untrusted, LLM-generated code safely within Go and Linux ecosystems.
• Deterministic Agent Architecture: Architectural patterns for multi-agent reasoning, state verification, and deterministic execution loops that eliminate hallucination cascades in production.
• Hardware-Software Co-Design & Reliability: Translating physical constraints (transients, thermal limits, control loop stability) into guardrails for automated systems.

What to expect

No generic AI hype or abstract speculation. Every post delivers either:

1. An architectural evaluation weighing latency, cost, and blast radius.
2. A concrete design workflow or failure analysis from simulation or lab testbenches.
3. A reproducible pattern you can evaluate and implement in your own systems.