Charles Hufnagel
Platform Architect, AI and Observability/Philadelphia
I build and operate the platform underneath production AI systems. At RealPage I am development lead on ARC, a portfolio of multi-agent services running on Google Kubernetes Engine that automate implementation and onboarding workflows. My work centers on the substrate rather than any one agent: a clone-and-go service factory that stands up a new agentic service with its CI/CD, per-environment deployment, and observability baseline already wired in, along with the LLM tracing and token-cost reconciliation standards that let the whole portfolio be measured the same way.
Before RealPage I led the Asia-Pacific team at STRATIS IoT, working on a telecom-grade, low-latency platform for shared smart-building gateways. Roughly a decade of engineering in total, most of it spent close to Kubernetes, infrastructure as code, and the question of how you can tell whether any of it is actually working.
What I work on
- Production agentic AI Tiered multi-agent orchestration with LangGraph and Google's Agent Development Kit, the A2A protocol, human-in-the-loop review and exception routing, evaluations and guardrails.
- Kubernetes and cloud platform GKE, EKS, and on-prem RKE2. Terraform and GitOps, Workload Identity Federation, Gateway API, secrets management, multi-cloud environment design.
- Observability OpenTelemetry instrumentation and the LGTM stack (Loki, Grafana, Tempo, Mimir), plus LLM observability: tracing, evaluation, and cost attribution across model providers. Elasticsearch and OpenSearch on the other side of that, including making the case for LGTM as an alternative to the Elastic Stack in a large Kubernetes environment.
- Agent tooling and developer experience MCP servers and reusable agent skills, model routing across providers (Gemini, Claude, OpenAI, and open-weight models), and the work of putting agentic coding tooling in front of a large engineering organization with cost controls and governance that hold up.
- Languages Python and Node.js, day to day.