Agent School
A set of runnable, graded example agents for the telecom industry, built to run on Red Hat OpenShift AI, showing how to turn a proven experiment into a governed agent workload.
Agent School provides agentic AI implementation examples for telecommunications, rebuilding proven experiments from a related project as governed agent workloads on Red Hat OpenShift AI. Each example follows the same architecture: the model is a stateless inference service, a harness owns the agent loop and tool calls, and a sandbox runtime controls what the agent is allowed to touch.
The five graded examples progress from a laptop-runnable network operations center assistant using a custom OpenAI-compatible loop, through a root-cause investigation agent and a fraud-triage agent built with LangGraph that adds a classic ML model as a governed tool with a human-approval gate, to closed-loop network operations and energy optimization agents that run live against two reference clusters. Each example can run offline against a scripted, time-snapshotted dataset with no model endpoint needed, or live against any OpenAI-compatible endpoint.