Senior DevOps Engineer, AI Platform

FloQast's AI products have outgrown the infrastructure patterns the rest of the platform runs on. Transform, AI Matching, and AutoBuilder are customer-facing products carrying real accounting workloads, and they behave nothing like a CRUD service. They call foundation models in multiple regions, execute generated code in sandboxes, spend money per token rather than per request, and fail in ways a 500-rate dashboard never catches.

Today, DevOps engineers carry this work alongside the wider fleet. We are making it someone's whole job. You will embed with the Transform and Close AI pods and own the AI runtime the way our other embedded DevOps engineers own their business unit's platform.

This is a DevOps role with an AI infrastructure specialization, not a research or modeling role. You will not train models or tune prompts for accuracy. You will make the systems that serve them fast, observable, multi-region, cost-bounded, and auditable.

Model serving: AWS Bedrock and Bedrock AgentCore across US, EU, and AU regions; TrueFoundry as the model-serving and deployment path

Sandboxed execution: AgentCore code-interpreter sessions with session lifecycle limits, network controls, and least-privilege IAM

AI observability: Grafana AI agent observability: token spend, per-model and per-region latency, throttle and retry rates, tool-call failures, and end-to-end agent traces, tied to journey-based SLOs

Delivery gates: model and prompt changes versioned, gated on eval and regression suites in CI (GitHub Actions), and rolled out or rolled back with Harness feature flags

Runs on: multi-region AWS (ECS Fargate and Lambda), defined entirely in Terraform

Days 1–30, map and instrument. Inventory the AI runtime across all three products: regions, model dependencies, IAM posture, IaC coverage, observability gaps. Establish an honest AI cost baseline attributed by business unit. Ship one visible observability win, such as a token-spend-and-throttle dashboard.

Days 31–60, close the highest-risk gap. Bring the multi-region model runtime fully under Terraform with no drift. Publish draft journey-based SLOs for the three products, with the pods bought in. Land one delivery-safety improvement: the matching eval suite running as a required CI gate.

Days 61–90, make it durable. Own the model-serving path end to end: production-ready, documented, and with no single point of knowledge. Write down the AI runtime patterns and runbooks so the next team extends them instead of rebuilding. Propose the next quarter of AI infrastructure work with the reliability or cost impact attached.

1. The multi-region AI runtime is production-grade: deployed and operated across US, EU, and AU, entirely in code, with no drift attributable to AI infrastructure.

2. Journey-based SLOs are live for Transform, AI Matching, and AutoBuilder, with AI-specific signals (tokens, throttles, tool-call failures, generation success) on the dashboards leadership already reads weekly.

3. AI spend is attributed and bounded: per-business-unit attribution, a defensible unit-cost metric, and a delivered reduction on the AI line.

4. Model and prompt changes ship behind gates: eval suites run in CI, changes are flag-controlled and reversible, and silent accuracy regressions are caught before production.

5. Tenant isolation is enforced and evidenced on every AI path, and the AI stack clears its SOC 2 / ISO cycle with evidence ready.

6. The runtime patterns are documented and adopted rather than rebuilt, with no single point of knowledge.

About four hours of conversation after a 30-minute recruiter screen. We move fast between stages.

1. Recruiter screen (30 min). Scope, level, location, compensation.

2. Hiring manager (45 min). The AI infrastructure problem in your own words; what you've owned and what broke.

3. Technical deep dive (75 min). A production incident or migration you led, interrogated properly, then hands-on: read an unfamiliar Terraform module and service, diagnose a described failure, propose the change. Use your normal tooling, including AI assistants; we care about how you verify.

4. AI infrastructure design (60 min). Design the serving, scaling, observability, and cost-control path for an LLM-backed feature under multi-region data-residency constraints. Whiteboard, no coding trivia.

5. Team panel (60 min). An engineering lead from Transform or Close AI, a Security or Compliance partner, and a conversation against FloQast's engineering values: Real Artists Ship; Do What Makes the Beer Taste Better; Cross the Water Then Build a Bridge; Look Before You Lock.