Principal AI Platform Engineer

Principal AI Platform Engineer

Choosing Capgemini means choosing a company where you will be empowered to shape your career in the way you'd like, where you'll be supported and inspired by a collaborative community of colleagues around the world, and where you'll be able to reimagine what's possible. Join us and help the world's leading organizations unlock the value of technology and build a more sustainable, more inclusive world.

New Co is a new AI-native product organization within Capgemini Financial Services. We build products, not projects: software for insurance claims, payment operations, and health operations, sold to banks, insurers, and health plans. Three product lines run on one shared platform, built by a deliberately small, senior team. Our engineering model is agentic: engineers author the specifications, tooling, evaluation suites, and guardrails, and AI agents do most of the implementation.

Humans own every consequential decision, and in our regulated domains some decisions are human-only by design.

Three product lines, one platform. You will own the platform that Claims, Payments, and Health run on: the agentic AI floor (model gateway, agent runtime, evaluation infrastructure, guardrails) and the shared product services around it (case management and work queues, integration connectors, multi-tenancy, metering). You run the platform as a product whose customers are our product teams, and you are its first and most senior engineer-leader.

Every hour of claims handling or payment processing our products automate rests on infrastructure your group builds.

The strategic vision, roadmap, and end-to-end lifecycle of the platform: from the model gateway and agent runtime to shared workflow, tenancy, and metering services. A competitive engineering strategy at the frontier: you track research and model releases as they land, decide what the platform adopts versus builds, and keep our capability curve ahead of what clients could assemble themselves. The closed improvement loops: production signals and evaluation verdicts feed reinforcement learning and fine-tuning pipelines that produce our own LLMs and SLMs;

product loops run automated end to end, with humans holding the gates. Build-vs-buy decisions across open-source and commercial AI infrastructure, and the boundary between what the platform provides and what product lines build themselves.

A track record leading platform or infrastructure teams that ran production systems for multiple product teams, with accountability for adoption, not just delivery. Hands-on credibility in modern AI infrastructure: LLM inference and serving, model gateways, vector search, guardrails, and evaluation systems. Frontier research fluency: you read post-training, reinforcement learning, and agentic-systems work as it lands and can turn it into engineering strategy;

an engineer who reads research, not a researcher at engineering distance. Cloud platform depth (AWS, Azure, or GCP) with Kubernetes and infrastructure-as-code at production scale. Experience delivering in a regulated industry, ideally financial services, or demonstrable fluency in what model-risk and security review requires of a platform. A platform-as-product mindset: you can talk about golden paths, voluntary adoption, and developer research as naturally as architecture.

Daily, hands-on use of AI coding assistants in your own work.

You have owned both an AI platform floor and shared business services (workflow, tenancy, billing/metering) in one charter. You have taken a model through post-training (RLHF, RLAIF, fine-tuning, or distillation to smaller models) into production. Published or open-source work in agent infrastructure or evaluation tooling. Cost management (Fin Ops) experience for LLM workloads. Financial services domain depth: you have shipped production systems for banks, insurers, or payment providers.

The reference technology stack for this role is our supported paved road: self-hosted Lang Smith and Lang Graph Platform as the agent runtime and evaluation plane, model providers behind a swappable gateway seam, PostgreSQL with pg vector plus Click House and S3-compatible object storage as the data platform, Neo4j Enterprise as the semantic knowledge graph, an agent memory plane serving episodic and precedent memory over MCP, MCP-native connectors, Open Telemetry and Grafana for observability, all on CNCF-conform ant Kubernetes with Helm and Argo CD, deployable to any hyper scaler or on-prem.

A tool-for-tool match is not expected: analogous experience counts fully.

Engineers write specs, harnesses, evals, and guardrails; AI agents execute the implementation loops. Review, not typing, is where engineering judgment goes. Three human gates govern everything we ship: spec approval, merge, and release. Regulated code paths (money movement, authentication, cryptography, secrets) are always human-owned. Small…