AI Pod - Cloud and Platform Engineer

Svitla Systems Inc. is looking for a Cloud and Platform Engineer for the AI Pod for a full-time position (40 hours per week) in LATAM.

Svitla’s Forward Deployed AI Engineering Pod is a small, senior team that works inside our client’s environment to take the client’s AI workflows into production one Workflow at a time, measuring the business result of each. The client pays a fixed monthly fee for working capability, not hours or headcount, and directs which Workflow the Pod works on. Each Workflow runs through Discover, Foundation, Build, Prove (shadow mode against live outcomes), and Operate (handover with runbooks, evaluation harness, and training), adapted from CRISP-DM, with the client’s own engineers shadowing throughout.

The Pod has up to five people in any month. Three Continuous Roles stay with the client for the whole engagement so that accountability never changes hands: the Outcome Lead, the Forward Deployed Technical Lead, and the Cloud and Platform Engineer. Specialist Bench roles (for example, agent and MCP engineering, data and semantic engineering, integration, document AI, evaluation, security architecture, data science, and adoption) rotate in as the active Workflow requires. The initial term is three months, then month to month for as long as the backlog holds Workflows the client wants in production. Continuous Role people are Svitla’s reusable core: they carry what they learn to the next Pod, reuse Svitla’s connectors, evaluation harnesses, governance patterns, and deployment templates, and improve them.

As the Cloud and Platform Engineer, you will own identity, deployment, and operations for the Pod. You will stand up and run the environments, infrastructure as code, pipelines, secrets, least-privilege access, observability, audit logging, and cost visibility inside the client’s cloud, and you will secure the early security and DevOps sign-offs that let a Workflow move from shadow mode to production without surprises.

Requirements

• Experience running ML or LLM workloads in production on at least one major cloud (AWS, Azure, or GCP), including deployment pipelines, monitoring, and incident handling.
• Strong experience with infrastructure as code and CI/CD (Terraform or equivalent, GitHub Actions or equivalent, and containers and Kubernetes or a managed equivalent), with examples of environments built from scratch.
• Hands-on experience with identity and access management: IAM policies, service accounts or workload identities, secrets management, and least-privilege design, with an audit or compliance review passed.
• Working knowledge of a modern data platform (Databricks, Snowflake, BigQuery, or similar) and of how governed access (catalogs, row and column security) is enforced.
• Experience with observability for AI systems: tracing, logging, evaluation metrics, and cost telemetry, not only uptime.
• Enough Python and scripting to automate the platform and to read the Pod’s application code.
• A security mindset: threat modeling, network boundaries, data classification, and encryption as default settings rather than add-ons.
• Clear, concise runbooks and handover documentation.

Will be a plus

• Experience with Google Cloud with Databricks, Azure OpenAI, or AWS Bedrock.
• Experience deploying and securing MCP servers, vector databases, or agent runtimes.
• FinOps practice and cost optimization for AI workloads.
• Experience preparing SOC 2, ISO 27001, HIPAA, SOX, or PCI evidence.

Responsibilities

• Stand up the Pod’s environments in the client’s cloud during Month 1: repositories, CI/CD, and dev, shadow, and production paths, all as infrastructure as code that the client keeps.
• Implement least-privilege access for people and agents: every agent is an identity with its own credentials, scopes, and joining and leaving controls, and reaches data only through the governed semantic layer and MCP surface.
• Own secrets management, key rotation, and the no-data-leaves-the-environment rule, including model provider and third-party API access.
• Build reproducible pipelines for data, model, prompt, and agent artifacts, with versioning and rollback.
• Put observability in from the first increment: tracing (for example, LangSmith or OpenTelemetry), evaluation and drift signals, audit logging of every agent action, and dashboards that the client’s operations team can read.
• Make run cost visible: model, API, compute, and storage cost per Workflow, reported monthly for the Value Ledger.
• Get the early security, compliance, and DevOps sign-offs from the client (architecture review, access review, and SOC 2 or regulatory evidence) so that the Prove gate is not blocked by a late finding.
• Write the runbook and the operations handover for each Workflow, and train the client’s team to run it.
• Maintain and improve Svitla’s reusable deployment templates, governance patterns, and observability packs, and bring them to the next Pod.