Platform Architect

Role: Platform Architect

Location: New York, NY (Hybrid, 3 days onsite per week)

Hire Type: contract

Job Description-

The most senior engineer on the team and the technical anchor for the platform. You design the reference architecture and personally build the hardest pieces — the lifecycle registries, the AI control-plane services that plug into the enterprise gateway, and the harness/memory services — setting the patterns the rest of the pod extends.

What You Will Do-

• Own the paved road — the reference architecture and the reusable registration pattern behind every asset (registration schema, semantic metadata, I/O contract, execution binding, governance constraints) so the agent, model, tool, workflow and skill registries all share one shape.
• Personally build the hardest pieces — the lifecycle registries; the AI control-plane services that plug into the enterprise gateway (registries, PII/PHI and prompt-injection scanning as a gateway policy, token-aware metering and multi-vendor cost attribution, AI trace/observability, and the multi-provider model-abstraction layer behind its fail-over); and the harness/memory & skills services.
• Build the agent-building paved road — the scaffolds, patterns and sub-agent topologies the team uses to stand up new agents fast, including the end-to-end builder agent (architect → provision → implement → test sub-agent → deploy → post-process → log).
• Deliver platform features as agents where it fits (e.g. a lifecycle-management agent, a skills-planning agent) — so the platform builds and operates itself, not just exposes CRUD APIs.
• Enforce portability in code — containerize everything; standardize lineage on MLflow + metadata, telemetry on OpenTelemetry, table formats on Delta UniForm / Iceberg, and open-weight serving on vLLM/Ray/DeepSpeed.
• Set the engineering bar — testing, CI/CD, IaC and security-by-default standards; review the team's hardest designs and pull requests; mentor the senior engineers.
• Build with AI and build agents end to end — like everyone on the team.

What We're Looking For — Required Qualifications-

• Principal-level full-stack — you build production services end to end (backend, APIs, and enough frontend to ship the registry and dev UIs) and you own systems in production.
• Distributed systems & platform engineering — Kubernetes, containers, IaC (Terraform), CI/CD, secrets/RBAC, and multi-tenant service design.
• Cloud depth with a portable mindset — strong on GCP (GKE, Vertex AI, IAM, VPC Service Controls, BigQuery), but standards-first by instinct.
• Hands-on GenAI / agentic engineering — LLM and agent runtimes, multi-agent and sub-agent orchestration, A2A and MCP/tool integration, retrieval/RAG, memory systems, and end-to-end builder agents.
• Security & governance by design — identity-aware access, PII/PHI handling, runtime guardrails, gateway/policy-as-code, and audit/observability.
• AI-assisted engineering — fluent and effective with AI coding tools (Cursor, Claude Code, Copilot, Windsurf or equivalent), and able to define the patterns and review discipline the team uses with them.
• Strong software engineering background — strong Python (and typically one of Go / Java / TypeScript).

Ideal Traits-

• Builds reusable, well-documented platform primitives that other engineers want to build on.
• Systems thinker; comfortable with ambiguity and incomplete requirements.
• Pragmatic about AI hype; reliability- and value-focused.
• Generous mentor; raises the bar by example.

Why This Role-

• Maximum technical autonomy — you own the hardest, highest-leverage parts of a brand-new platform and the patterns everything else inherits.
• Staff-level impact without management overhead — a first-class IC track, paid at the top of the market, anchoring a team of strong engineers.
• Genuinely hard problems — AI control planes, multi-agent governance, enterprise memory, and builder agents done right.

Success Looks Like-

• Reusable paved roads (registries, harness, builder agents) the whole team builds on.
• The hardest platform pieces shipped and reliable in production.
• A high, consistent engineering bar across the pod.