AI Platform Engineer

Capgemini•Atlanta, GA

3d•$129,000 - $143,000

About The Position

The AI Platform Engineer builds and operates shared AI platform components end to end. Components may include a model gateway for routing, failover, caching, and per-line cost attribution; agent runtime and orchestration; evaluation harnesses; guardrail engines; tool and agent registries; and shared product services. The role also develops training and adaptation pipelines that turn production traces and evaluation verdicts into reinforcement-learning and fine-tuning datasets, and infrastructure to train, evaluate, and serve tuned large and small language models alongside vendor models. The engineer owns the reliability, latency, and cost of their platform services and builds self-service components with documentation and sensible defaults. They collaborate closely with product teams, initially embedding with a product line to build shared services and moving proven services into the platform. The reference stack includes self-hosted LangSmith and LangGraph Platform, swappable model providers, PostgreSQL with pgvector, ClickHouse, S3-compatible storage, Neo4j Enterprise, MCP-based agent memory and connectors, OpenTelemetry, Grafana, and CNCF-conformant Kubernetes with Helm and Argo CD. Equivalent experience with comparable tools is accepted. The team is small and senior by design. Engineers write specifications, harnesses, evaluations, and guardrails, while AI agents execute implementation loops. Human approval gates apply to specifications, merges, and releases; regulated code paths such as money movement, authentication, cryptography, and secrets remain human-owned. The position offers a posted base compensation range of 129,000 to 143,000. Additional variable incentives, bonuses, or commissions may be available depending on the position and applicable laws. Regular full-time employees may be eligible for a comprehensive benefits package subject to local policy and eligibility.

Requirements

Strong production engineering skills with Python and one systems language; Go or Rust are welcome.
Experience with Kubernetes, infrastructure-as-code, and one major cloud.
Hands-on experience with LLM infrastructure, including a model gateway pattern, model serving, and vector or retrieval systems.
Experience building with an agent orchestration framework and understanding of tool calling and the Model Context Protocol (MCP).
Experience building or operating evaluation harnesses, including golden datasets, CI regression gates, and LLM-judge calibration, with the ability to explain why evaluations gate merges.
Post-training experience, such as fine-tuning or post-training a model, building the supporting data pipeline, or reproducing recent research techniques in production systems.
Daily hands-on use of AI coding assistants in the development workflow.
Depth in four or five relevant areas is valued more than surface familiarity across all areas.
Equivalent experience with comparable tools and components is accepted; an exact match to the reference stack is not expected.

Nice To Haves

Experience with guardrail engines or AI observability, such as NeMo Guardrails, OpenTelemetry GenAI conventions, LangSmith, Braintrust, or equivalent.
Reinforcement-learning infrastructure experience, including reward modelling or RLHF- or GRPO-class pipelines, or SLM distillation experience.
Multi-tenant SaaS infrastructure experience, including tenancy isolation, metering, or usage-based billing.
Financial-services engineering experience in banking, insurance, or payments; experience in another regulated industry or security engineering also counts.
Open-source contributions or a portfolio of deployed agents and evaluation suites.

Responsibilities

Own one or more AI platform components end to end, such as a model gateway, agent runtime and orchestration, evaluation harness, guardrail engine, tool and agent registry, or shared product services.
Build training and adaptation pipelines that convert production traces and evaluation verdicts into reinforcement-learning and fine-tuning datasets.
Develop infrastructure to train, evaluate, and serve tuned LLMs and SLMs behind the same gates as vendor models.
Maintain the reliability, latency, and cost baselines of platform services.
Build self-service platform components with documentation and sensible defaults for product engineers.
Co-build new shared services with product teams, embedding with a product line and graduating proven services to platform services.
Write specifications, harnesses, evaluations, and guardrails, and review AI-agent-produced implementation work.
Apply human approval gates for specifications, merges, and releases; keep regulated code paths human-owned.

Benefits

Paid time off, with vacation of 12–25 days depending on employee grade and policy.
Company-paid holidays.
Personal days.
Sick leave.
Medical coverage, or provincial healthcare coordination in Canada.
Dental coverage.
Vision coverage.
Retirement savings plans, such as a 401(k) in the U.S. or RRSP in Canada.
Life insurance.
Disability insurance.
Employee assistance programs.
Other benefits as provided by local policy and eligibility.