AI Engineer at Gen · Czech Republic
Multi-agent orchestration, LLM evaluation and grounded retrieval. Designed, coded and shipped at enterprise scale.
› trace complete · citations attached · output streamed
Built, not theorised
01 / Production Evidence
Škoda Auto · Enterprise AI
I architected, coded and led an enterprise multi-agent assistant used across business units. The system combined a forked Google ADK, custom flows, typed streaming, tool integrations and long-lived evaluation.
internal users
designed and coded
evaluation over time
02 / Interactive System Model
Inspect the decisions behind a production agent platform. The diagram is illustrative; the engineering principles are applied in practice.
Request flow · illustrative sequence
Follow a request through the system.
Layer 01
A bounded orchestrator classifies intent, selects a specialist flow and keeps tool permissions explicit. Not every problem needs an autonomous agent.
Layer 02
Retrieval and typed tool calls bring approved data into context. Structured outputs keep downstream systems in control.
Layer 03
Golden datasets, trace-level checks and regression benchmarks turn AI quality into an engineering signal.
03 / Selected Work
Public code and detailed system designs. Each study states assumptions, weighs trade-offs and commits to a decision.
A governed multi-use-case agent platform for a regulated operator, covering use-case selection, lifecycle gates, architecture and delivery.
Read the case study ↗A grounded intent-to-catalog pipeline designed for 1.2 million conversations a day within a strict latency budget.
Read the design ↗Self-hosted RAG chatbots from your PDFs, embeddable with one script tag using your own keys and data.
A local-first transcription and dubbing studio designed to keep confidential media on your machine.
04 / Engineering Profile
I work where LLM behaviour meets real software constraints: permissions, latency, observability, maintainability and changing models. I have five years of hands-on engineering experience across AI, backend and architecture.
Currently an AI Engineer at Gen. Previously AI Engineer and Tech Lead at Škoda Auto.
Agent orchestration, Google ADK, MCP, RAG and contextual memory
Golden datasets, LLM evaluation, regression testing and observability
Python, Node.js, C#/.NET, Kotlin/JVM and event-driven systems
05 / Quick Answers
Production multi-agent systems, LLM evaluation frameworks, grounded RAG, tool integrations and reusable AI platform architecture.
ŠkoPilot at Škoda Auto: an enterprise multi-agent assistant serving more than 30,000 internal users. I owned architecture, implementation and technical leadership.
Yes. The case studies document architectural decisions and trade-offs. DocuSage and Dabuj are MIT-licensed public repositories.
Yes, especially lead or architect-level work in AI systems, evaluation and developer tooling. Find me on LinkedIn or explore my code on GitHub.
Build the next reliable AI system
Open to thoughtful conversations about production AI, platform architecture and technical leadership.