> Markdown version of [/jobs/ext/3740073-senior-ai-platform-engineer](https://www.wearedevelopers.com/jobs/ext/3740073-senior-ai-platform-engineer). Every page supports `.md` or `Accept: text/markdown`. Links point to the HTML versions so they work for humans too. Agent guide: [/agents.md](https://www.wearedevelopers.com/agents.md). --- # Senior AI Platform Engineer - **Company:** gb Capgemini Uk Plc (FS) - **Location:** UK (Remote available) - **Experience:** Expert - **Contract:** Permanent contract - **Skills:** LangGraph Framework, Artificial Intelligence, Amazon S3, Automated Storage and Retrieval Systems, Software as a Service, Data Infrastructure, Graph Database, Python (Programming Language), PostgreSQL, Neo4j, Octopus Deploy, Reinforcement Learning, Large Language Models, Grafana, NeMo Guardrails, AI Coding Agents, Agentic-AI, Pgvector, AI Platforms, LangSmith, Kubernetes, Free and Open-Source Software, LiteLLM, Vertica, Model Context Protocol, Data Pipelines - **Published:** October 15, 2026 - **Apply:** https://www.totaljobs.com/job/senior-ai-engineer/capgemini-uk-plc-fs-job108173107 ## About the Role * Strong production engineering: Python plus one systems language (Go or Rust welcome), Kubernetes, infrastructure-as-code, and one major cloud * Hands-on experience with LLM infrastructure: a model gateway pattern (such as LiteLLM or in-house), model serving (such as vLLM or managed endpoints), and vector or retrieval systems * Agent-systems experience: you have built with an agent orchestration framework (such as LangGraph or first-party agent SDKs) and understand tool calling and the Model Context Protocol (MCP) * Evaluation engineering: you have built or operated eval harnesses (golden datasets, regression gates in CI, LLM-judge calibration) and can explain why they gate merges * Post-training fluency: you have fine-tuned or post-trained a model, built the data pipeline behind one, or reproduced techniques from recent papers in production systems * Daily, hands-on use of AI coding assistants as part of your own development workflow, * Guardrail-engine or AI-observability experience (NeMo Guardrails, OpenTelemetry GenAI conventions, LangSmith, Braintrust, or equivalent) * Reinforcement learning infrastructure (reward modelling, RLHF or GRPO-class pipelines) or SLM distillation experience * Multi-tenant SaaS infrastructure: tenancy isolation, metering, usage-based billing * Financial services engineering (banks, insurers, or payment providers) is the ideal background; other regulated-industry or security engineering depth also counts * Open-source contributions or a portfolio of deployed agents and eval suites (we would rather see this than a CV) The reference stack ## Description You will build the platform floor that our AI products for banks and insurers run on: the model gateway every inference call passes through, the agent runtime that executes planning loops and tool calls, the evaluation infrastructure that gates every release, the guardrail engine, and the shared services (case management, connectors, tenancy, metering) that stop three product teams building the same thing three times. This is production infrastructure for regulated industries, built by a small senior team with heavy AI leverage. The ambition runs past serving frontier models: we close the loop from production feedback through reinforcement learning and fine-tuning, and train our own LLMs and SLMs where evaluations and economics justify it. What you will own * One or more platform components end to end: model gateway (routing, failover, caching, per-line cost attribution), agent runtime and orchestration, evaluation harness, guardrail engine, tool and agent registry, or shared product services * The training and adaptation loop: pipelines that turn production traces and evaluation verdicts into reinforcement learning and fine-tuning datasets, and the infrastructure to train, evaluate, and serve NewCo-tuned LLMs and SLMs behind the same gates as vendor models * Reliability, latency, and cost of what you build; platform services carry baselines and you hold them * The self-service surfaces product engineers use: your components ship with documentation, sane defaults, and no ticket queue * Co-building with product teams: new shared services start embedded with a product line and graduate to platform services when proven What you will need, The reference technology stack for this role is our supported paved road: self-hosted LangSmith and LangGraph Platform as the agent runtime and evaluation plane, model providers behind a swappable gateway seam, PostgreSQL with pgvector plus ClickHouse 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, OpenTelemetry and Grafana for observability, all on CNCF-conformant Kubernetes with Helm and Argo CD, deployable to any hyperscaler or on-prem. A tool-for-tool match is not expected: analogous experience counts fully. If you have built and operated systems of this shape on comparable components (a different orchestration framework, graph engine, evaluation platform, or serving stack), you have what we are looking for. How we work * 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 and senior by design. No separate QA function, no scrum masters; quality comes from evaluation gates and whole-team review rituals. * Domain experts (claims practitioners and payment scheme experts) are full-time members of the product teams you will serve. Success in year one * A platform component you own is consumed self-service by at least two product lines, with measured adoption * Evaluation gates you built block real regressions before clients ever see them * Per-line cost attribution from the gateway is trusted enough to drive planning * Production traces and evaluation verdicts flow into curated adaptation datasets, and a NewCo-tuned model serves traffic behind the same gates as vendor models We are a Disability Confident Employer Capgemini is proud to be a Disability Confident Employer (Level 2) under the UK Government's Disability Confident scheme.As part of our commitment to inclusive recruitment, we will offer an interview to all candidates who: * Declare they have a disability, and * Meet the minimum essential criteria for the role.