Shipping Agentic Systems
in production
I architect the foundational infrastructure for AI-native applications — LLM routing, agent observability, and the cloud platforms beneath them — backed by 8+ years leading teams from business strategy to delivered software.
What I build
Not a list of tools — the systems I own end-to-end, from infrastructure to the interface on top.
AI-Native Infrastructure & Tooling
The substrate beneath AI products: MCP servers, RAG pipelines and vector search, plus the developer tooling on top — agent-first CLIs, sandboxed environments, and secure code execution.
Agentic Systems & Copilots
Multi-agent orchestration across the spectrum — autonomous long-horizon workers, customer-facing copilots, and internal assistants. Tool-use, human-in-the-loop guardrails, and context engineering wired through MCP.
Observability & Telemetry
Hierarchical traces that capture every LLM call, tool invocation, and retrieval step — filterable by user, session, cost, and latency. LLM-as-a-judge evaluation, prompt management, and golden datasets on a high-throughput telemetry engine.
Cloud & DevOps
Infrastructure-as-Code across AWS, GCP, and Azure, with Kubernetes at scale. DevOps pipelines with AI at their core — automated PR reviews and self-healing, self-building systems on GitHub Actions and GitLab CI.
High-Performance Backends & Intuitive Frontends
End-to-end product engineering, backed by 8+ years full-stack: Go and Python services over real-time data layers, paired with intuitive React / Next.js frontends — including generative UI — built for throughput and heavy user loads.
Leadership & Communication
Building effective multidisciplinary teams and leading them from business strategy to delivery. Fluent translating between technical and business contexts, with a genuine focus on mentoring, teaching, and growing engineers. Owns enterprise trust end-to-end: ISO 27001, SOC2, GDPR and EU AI Act.
Projects I've shipped
The path so far
Select a node to inspect the role. [ click to expand ]
Works across the stack
From the intelligence layer down to the platform it runs on — no framework lock-in.