Architecture & Leadership

Zoltán Takács

Setting the technical strategy and architectural direction for enterprise-wide application and cloud security. Pioneering the intersection of AI and InfoSec by integrating LLMs into core security workflows: engineering AI-driven alert triage, building secure-by-default tooling, and designing hardening patterns for Kubernetes and multi-cloud (GCP/AWS) environments. Translating recurring root-cause vulnerabilities into systemic, organization-wide prevention mechanisms.

role: Staff Security Software Engineer, InfoSec R&D certified: GCP Professional Cloud Architect & Security Engineer stack: LangGraph, LLM Agents, Kubernetes, GCP, AWS, Terraform

$ career

Site Reliability Engineering Manager

Architected and deployed DevSecOps pipelines. Partnered across PaaS and SaaS product development cycles to ensure technically feasible, high-reliability deployments. Mentored SREs, streamlined new product onboarding, and engineered out operational toil through automation.

DevSecOps PaaS/SaaS SRE Automation
Technical Manager / Principal Software Engineer

Directed a team of principal software engineers to resolve high-stakes technical escalations. Delivered Cloud Native solutions for the Data team, optimizing complex infrastructure workflows. Served as a global mentor for Site Reliability Engineers, standardizing technical excellence across international teams.

Cloud Native Data Workflows SRE Mentorship Escalations
Principal Software Engineer, AI Integration

Spearheaded the integration of Vertex AI models into internal tools and services, fundamentally optimizing support workflows. Architected cloud-native applications utilizing TypeScript for backend APIs and Python for model interaction. Deployed Terraform for IaC to provision, scale, and secure enterprise AI infrastructure.

Vertex AI TypeScript Python Terraform GCP
Staff Security Software Engineer, InfoSec R&D

Setting the technical strategy and architectural direction for enterprise-wide application and cloud security. Pioneering the intersection of AI and InfoSec by integrating LLMs into core security workflows: engineering AI-driven alert triage, building secure-by-default tooling, and designing hardening patterns for Kubernetes and multi-cloud (GCP/AWS) environments. Translating recurring root-cause vulnerabilities into systemic, organization-wide prevention mechanisms.

LangGraph LLM Agents Kubernetes GCP AWS Terraform

$ cat education.md

MSc in Computer Science and Engineering
Thesis: GPU-based signal processing methods for turbulence signal analysis support
  • Developed parallel signal processing algorithms using the NVIDIA CUDA architecture to accelerate Fast Fourier Transform (FFT) power spectrum calculations for massive datasets.
  • Collaborated with Wigner FK RMI to further develop and optimize a module within the Fluctuation IDL Program Package (FLIPP).
  • Achieved up to a 27x performance speedup on GPU compared to traditional CPU processing.
NVIDIA CUDA FFT GPU Computing Signal Processing Parallel Algorithms

$ cat ai_engineering_principles.md

Cost-Aware Engineering

Every model call has a price tag. Token-level attribution, accelerator economics (GPU vs. TPU), and workload-specific cost governance are not afterthoughts. They are architectural requirements that must be designed in from day one.

AI Is Software. Engineer It Like One.

Models without observability, error handling, and deployment pipelines are experiments, not products. AI features need the same engineering rigor as any production service: versioned, tested, monitored, and built to fail gracefully.

Multi-Directional Adoption

Top-down governance and cost control set the guardrails. Bottom-up experimentation and hands-on tool evaluation drive adoption. Neither direction alone produces sustainable AI integration across an engineering organization.

The Platform & AI Lab

~/ai-platform$

AI Integration & Governance

AI adoption requires a multi-directional approach: top-down strategic governance and cost control, combined with bottom-up hands-on experimentation, tool evaluation, and infrastructure testing. Neither direction alone is sufficient.

Enterprise AI on GCP

  • Multi-workload Vertex AI deployments with workload isolation and resource governance
  • Cloud accelerator economics: GPU vs. TPU trade-off analysis for training and inference workloads
  • Application-level AI API consumption tracking and cost attribution across business units
  • Model deployment pipelines with versioning, rollback, and A/B traffic splitting

Practical AI Engineering

  • 24h AI Hackathon: Built autonomous agent workflows using LangGraph for multi-step reasoning, tool invocation, and stateful conversation graphs under time constraints
  • Local AI Execution: Testing Gemma 4, Qwen, and Llama models on constrained hardware, profiling inference latency, memory footprint, and quantization trade-offs
  • Cost Governance: Token-level usage dashboards, per-team quota enforcement, and anomaly detection on API spend patterns
Vertex AI LangGraph Gemini RAG Agent Workflows Model Garden Gemma 4 Qwen Llama GCP GPU/TPU Cost Attribution
~/homelab$

Homelab Infrastructure

Applying enterprise-grade SRE practices, high reliability, and a secure-by-default mindset to personal infrastructure. Every service is containerized, reverse-proxied, and monitored, because production discipline doesn't stop at the office door.

Networking & Security

  • Custom domain management with A/AAAA/MX record verification and DNSSEC
  • Secure remote access orchestration via Caddy/Nginx reverse proxies
  • Automated SSL/TLS certificate management with zero-touch renewal
  • Controlled application exposure: principle of least privilege applied to network boundaries

Self-Hosted Platform

  • Immich: Privacy-first photo management. Google Photos alternative with on-device ML classification, zero cloud dependency
  • Jellyfin: Self-hosted media server with hardware transcoding, multi-client streaming, and granular library permissions
  • Docker Compose: Declarative infrastructure-as-code for all services, with health checks, restart policies, and volume persistence
Docker Caddy Nginx Immich Jellyfin DNSSEC SSL/TLS

PwC 24h AI Hackathon

~/hackathon$

Agentic Contract Review Workflow

In 24 hours at PwC Hungary's AI Hackathon, we built a fully functioning, risk-focused contract management system powered by LangGraph agentic workflows. AI agents handle drafting, verification, revision, and rendering of contract review outputs through a controlled loop, connected to contract intake, metadata extraction, organization-specific guidance, and low-code platform updates.

What We Built

  • Multi-agent contract review pipeline using LangGraph with stateful conversation graphs
  • Automated metadata extraction and organization-specific guidance injection
  • Human-in-the-loop review experience with controlled agent handoff points
  • Low-code platform integration for downstream workflow updates
  • Risk-focused contract analysis with structured output rendering

Stack

LangGraph Claude Agentic Workflows Contract AI Python Low-Code Integration
event PwC Hungary AI Hackathon
duration 24 hours
team Zsolt Balogh & Zoltán Takács
result Fully functional prototype

Certifications

Professional Cloud Security Engineer

Professional Cloud Security Engineer

Google ID: 67857912
Professional Cloud Architect

Professional Cloud Architect

Google ID: 67857912
Generative AI Leader

Generative AI Leader

Google ID: 67857912
Build with Vertex Technical Expert Badge

Build with Vertex Technical Expert Badge

Google Cloud ID: 67857912

7 Habits of Highly Effective People

FranklinCovey ID: x72ibdixoehx

Associate Cloud Engineer

Google Cloud ID: 26285051

Microsoft Certified: Azure Fundamentals

Microsoft ID: H615-8710

AWS Certified Cloud Practitioner

Amazon Web Services (AWS) ID: 370MGZP13NR4QZCG