Sr. SRE Platform Architect

Key Responsibilities

• Own the end-to-end architecture of the NeoCloud SRE platform — the substrate that observes, protects, and operates a multi-region GPU rental fleet across self-built and OEM-rented data centers.
• You are the single point of architectural accountability across the platform's ~57 bounded contexts, ~12 frameworks, and three operational tiers (Edge DC → Regional Controller → Global Hub).
• This role is for someone who writes the design, defends it under review, and shepherds it through the engineering squads that build it.

What You'll Do

• Write and maintain the platform architecture document — keep the design coherent across all sections, frameworks, and tiers.
• The current document is your starting point.
• Review every framework-level change — new bounded context, new plugin kind, tier-deployment shift, schema change, naming change, cross-context contract change.
• Architecture changes ride GitOps PRs like any other artifact.
• Set design invariants — residency rules (raw data stays in Region), Tier 2 self-sufficiency budget (≥ 24 h), survival-uplink contracts, naming conventions, SLO catalogues, redaction-at-boundary rules.
• Run the plugin framework — every extension uses one uniform contract (Common + Domain manifest, lifecycle, observability).
• You author and evolve this contract.
• Decide tier placement — what runs at Edge DC vs Regional Controller vs Global Hub, with data-residency / compliance / availability tradeoffs explicit.
• Coordinate with cloud-service teams and tenants — they author plugins, SDKs, dashboards, agent recipes that ride the platform. You set the contracts they consume.
• Coordinate with Security — joint ownership of vulnerability management, exposure management, joint operations. Security owns policy and risk acceptance; you own the operational mechanisms they ride.
• Pre-flight roadmap items — for any new capability, produce a one-page design that fits the existing layered model (L1–L6), tier topology, naming conventions, and extension contracts before implementation starts.
• Defend the design under review — say no to scope creep, special-case workarounds, and one-off integrations that don't fit the framework model. Say yes when a new plugin kind is genuinely needed.

Qualifications

• 10+years of production SRE / platform-engineering / infra-architecture, including ≥ 3 years at architect level.
• Hands-on with GPU / AI-compute infrastructure — NVIDIA GPU ops (DCGM, MIG, vGPU, NVLink/NVSwitch, XID semantics, NCCL), InfiniBand or RoCE fabrics (subnet manager, fabric partitioning, optical health), HPC storage (Lustre, NetApp/Pure/DDN/VAST, NVMe-oF).
• Multi-region observability at scale — metrics / logs / traces / profiles / analytics-lake substrate; recording rules, MWMBR burn-rate alerting, SLI/SLO discipline.
• Cluster platforms — first-hand experience with Kubernetes (control plane + GPU Operator + topology-aware scheduling) AND at least one of Slurm / Volcano / Kueue / Ray / KubeRay.
• Data-center operations — ZTP, BMC/IPMI/Redfish, BIOS/firmware lifecycle, RMA, multi-vendor OEM management (self-built + leased DC mix).
• Strong DDD instincts — bounded contexts, public contracts, no shared databases, one-context-one-repo discipline.
• Plugin framework design — you have built (or substantively contributed to) a real extension framework with a uniform manifest + lifecycle.
• Writing fluency — you can author and maintain a multi-thousand-line architecture document under review without it drifting; you can also write a one-pager an executive will read.
• Cross-team operating tempo — design reviews, runbook authorship, on-call shadowing, post-mortem facilitation
• Hyperscale or NeoCloud experience
• BS/MS in Computer Science or similar