Principal AI Platform Engineer

• As we continue to grow, we’re looking for a Principal AI Platform Engineer to be a foundational hire on our new AI Operations team. This is the engineer who builds the plumbing of our internal agentic workforce: the AWS hosting layer, the Cognito-backed identity model, the Accountability Registry that says who built what, the Agent Registry that governs who can act, and the Tool Catalog that lets our teams compose new capabilities safely

• You’ll partner with the CTO and a UX-focused builder to make Hi Marley an AI-native company from the inside — and you’ll do it on a strict AWS stack because our customers are insurance carriers who need real auditability, not a wrapper

• Build the AI Operations platform on AWS: Own the hosting, deployment, and infra-as-code layer for every internal agent and tool we run. Bedrock for models, Cognito for auth, CDK or Terraform for everything else. No third-party platforms — we are AWS-only by design

• Architect the registries and catalogs: Design and build the Accountability Registry (people, ownership, roles), the Agent Registry (identity, permissions, data scope, audit trail), and the Tool & App Catalog (discover, register, approve). These are the spine of agentic governance at Hi Marley

• Solve the identity-inheritance problem: An agent acts with the inherited permissions of the human who invoked it. You’ll design and ship the patterns — Cognito + IAM + scoped tokens — that make this real, deterministic, and auditable

• Build the developer experience for internal agent builders: Templates, SDK helpers, deploy paths, observability, cost tracking. Make it so that every Hi Marley team can ship their own agent without reinventing security or hosting

• Partner on security tooling with our Compliance & IT lead: Implement the scans, audits, MCP governance hooks, and DLP signals that the AI Security Ops role will use. You build the rails; they build the gatekeeper

• Evangelize the platform internally: Lunch-and-learns, internal docs, demo droplets, and “office hours” so other teams know what’s available and how to use it. Communication and product instinct are as load-bearing as the AWS depth

• Use AI in your daily flow: Cursor, Claude Code, agentic coding tools — your default question is “could an agent do this?” before you write the code yourself

Benefits

• Core values-based leadership

• A culture of employee engagement, diversity and inclusion

• A fun, lively startup culture

• Ample opportunities to learn and take on new responsibilities in a fast-paced, growth-mode startup

• Open vacation policy - we all work hard and take time for ourselves when we need it

• Full benefits package including parental leave, a matching 401k program, and medical, dental, vision, disability, and life insurance

• Generous stock options - we all get to own a piece of what we’re building- Strong product instincts, not just infra: You think about the developer experience of the people consuming your platform. You can design an SDK or a registry schema that other teams actually want to use

• Excellent communicator: You can present platform work to engineers, executives, and customers, and write docs people actually read. This role only succeeds if the rest of the company knows what’s available

• MCP / agentic systems literacy: Familiarity with Model Context Protocol or comparable agent-tool patterns. You don’t have to have shipped an MCP server, but you should be able to talk credibly about why MCP exists and what its security boundaries are

• Deep AWS expertise: You can architect a multi-tenant platform on AWS without needing to outsource the hard parts. Bedrock, Cognito, IAM, EventBridge, Lambda, DynamoDB or Postgres, plus IaC (CDK or Terraform). Bonus: AgentCore, SageMaker, Workshop Studio

• Architectural integrity over framework hype: You’d rather build the right primitive on AWS than reach for a “magical” higher-level abstraction. You can write Python, TypeScript, or Go fluently — the language matters less than the systems thinking

• AI-forward: You’ve shipped real LLM-backed systems in production — not just prototypes — and have opinions about token cost, latency, evaluation, observability, and where determinism matters vs. where it doesn’t

• Identity & authorization fluency: You’ve built systems where authorization is the architecture, not bolted on. OAuth 2.1, scoped tokens, role inheritance, least-privilege design — these are second-nature to you

• A genuine curiosity about AI and emerging technologies, paired with the judgment to apply them thoughtfully and responsibly