years across research, cloud, and startups
publications in networking and optimization
engineers aligned on complex delivery programs
Jun (Joel) He, PhD
Systems architect for governable AI infrastructure.
Joel builds the control-plane layer OpenKedge argues every serious AI system will need: a deterministic path between probabilistic reasoning and live operations.
His work combines academic depth in networking and complex system optimization with hands-on delivery across global cloud, startup, and research environments. The throughline is simple: autonomous systems need governance that survives real operational pressure.
Research discipline, production instincts, and a bias for systems that operators can actually trust.
OpenKedge is shaped by the gap Joel has seen repeatedly: AI can produce plausible operational intent, but infrastructure still needs deterministic authorization, scoped execution, and durable evidence. That gap is architectural, not cosmetic.
A control-plane view of AI infrastructure.
Distributed control planes
Designing stateful, policy-aware infrastructure that can coordinate safely across teams, regions, and failure domains.
Execution governance
Bounding AI actions with explicit context, authority, identity, and evidence before production systems change.
Operational delivery
Moving ambiguous 0-to-1 platforms from research-shaped insight into production systems with real operators and constraints.
"Systems must be designed to remain safe and controllable even when the organization operating them is not. The correctness of the AI agent is secondary; the correctness of the system governing it is primary."
This perspective comes from delivering complex infrastructure through unstable ownership, shifting priorities, and imperfect coordination while keeping the system accountable to operators.
Treat AI output as a request, not permission.
Make every high-impact action explainable before it executes.
Design for organizational drift, not perfect coordination.
Give operators evidence they can replay, audit, and challenge.
Useful where autonomy meets authority.
The founder perspective is strongest in environments where AI systems cannot be trusted with unchecked privilege, and where governance needs to be a first-class runtime concern.
Regulated cloud operations
Governed action paths for teams that need strong auditability, scoped execution, and explicit impact control.
National-scale infrastructure
Control-plane patterns for multi-organization environments where trust boundaries cannot be handled by a single API owner.
Autonomous systems strategy
Advisory work for leaders deciding where AI agents may reason, where they may act, and what must sit between those steps.
Evidence-backed governance
Architectures where policies, identities, decisions, and execution artifacts become part of a durable operational record.