AI Platform Engineer

• BSc or MSc in Computer Science, Software Engineering, AI, or related field.
• 4+ years in Software Engineering or Platform Development, with demonstrable recent experience in Generative AI and/or Agentic Systems.
• You don't need to tick every box. Strong Python + hands-on LLM application experience + a platform/developer-experience mindset matter most; we expect you to grow into the rest.
• AI / LLM engineering. Practical experience building LLM-powered applications; familiarity with RAG and agentic patterns (ReAct, plan-and-solve, multi-agent) and a clear sense of when not to use an autonomous agent.
• Evaluation & quality (core). Designing eval harnesses, golden tests and retrieval-quality metrics for LLM/RAG systems (grounding, retrieval precision, hallucination control); context engineering over model selection.
• Backend & API development. Python proficiency is highly desired (FastAPI, Pydantic, async); designing and operating production backend services and well-versioned APIs.
• Platform / Developer-Experience engineering. Building reusable, self-service components and paved roads (templates, SDKs, golden paths) and operating multi-tenant services in production (SLOs, observability, "you build it, you run it").
• Software development across the stack. Enough context across frontend, backend, and infrastructure to contribute across DevHub's stack (with AI-assisted coding) - no need to be a full-stack expert in every layer.
• Cloud & infrastructure. Hands-on Azure and containerization (Docker, Kubernetes/AKS); infrastructure-as-code (HCL/Terraform, modular).
• Data & state management. Relational/non-relational databases (PostgreSQL) and vector stores (e.g. Azure AI Search); managing context and state at scale.
• DevOps & production operations. CI/CD (Git, GitHub Actions), monitoring/observability, and security best practices in production.
• AI / LLM ecosystem. LLM providers/APIs (OpenAI, Anthropic, Mistral), managed AI services (Azure AI Foundry, Databricks), and the MCP standard. (At DevHub, models are consumed through the internal AI Gateway, not provider SDKs directly.)
• Security & multi-tenancy. Authentication/authorization, RBAC, tenant isolation, guardrails, auditability; awareness of the OWASP LLM Top 10.
• Agentic standards beyond MCP (e.g. A2A); spec-driven ("spec-kit") agentic development with AGENTS.md / skill conventions.

• Databricks / Unity Catalog - DevHub's core data platform; DevHub governs the Databricks account and environments (workspaces, Unity Catalog, blueprint lifecycle), not the Enterprise Data Lake or data distribution. Familiarity is a strong plus.
• TypeScript alongside Python; Grafana; MLflow and data pipelines.
• FinOps / cost attribution for AI features.
• Message queues / event-driven & data-intensive architectures (e.g. RabbitMQ, Azure Event Grid).
• Workflow engines / state machines; OpenTelemetry & distributed tracing. Experience with EnvoyProxy

• A secure work environment because your health, safety and wellbeing is always our top priority.
• Flexible work schedule and Home-office options, so that you can balance your working life and private life.
• Learning and development opportunities
• 25 holiday days per year
• 5 additional days (readjustment)
• A collaborative, trustful and innovative work environment
• Being part of an international team and work in global projects
• Relocation assistance to Madrid provided

We are seeking a hands-on AI Engineer for BASF's DevHub - the Internal Developer Platform (IDP) used by thousands of engineers and product teams across BASF. DevHub already ships an enterprise AI Gateway (50+ governed models, Entra ID, EU data residency, per-cost-center billing, Grafana observability) and a catalog that is a schema-validated knowledge graph of every product and its infrastructure. Your mission is to make AI a first-class platform capability: build reusable, production-grade AI services and developer experiences that help users discover, create, configure, operate, scale and govern their products, surfaced where they already work - the portal, the IDE (GitHub Copilot/MCP) and Teams. You will treat the platform as a product - shipping paved-road components other teams reuse, serving both humans and agents, with the multi-tenant scoping, cost-tracking, guardrails and governance an enterprise platform demands.

• Treat the platform as a product. Build paved roads and self-service: reusable AI building blocks (shared retrieval/"context engine," guardrail & evaluation libraries, an MCP/tool layer), scaffolder templates, SDK/API access and stable, versioned interfaces - built once, reused across features.
• Ship AI experiences that delight developers. Grounded, well-cited assistants, copilots and wizards across the product lifecycle (e.g. a conversational knowledge assistant over our docs and catalog), meeting users on the portal, IDE (Copilot/MCP) and Teams via one shared API.
• Serve humans and agents. Expose platform capabilities through an MCP / SDK / API surface - read-first, RBAC- and tenant-aware - so internal and external AI clients can query and (later, gated) act on the platform. See the AI-Assisted Platform Strategy RFC.
• Own evaluation and quality. Build eval harnesses, golden tests and retrieval-quality metrics so features are correct, grounded and regression-tested in CI; invest in context engineering over model-shopping - the Gateway already solves model choice.
• Pick the right pattern. Prefer deterministic pipelines + structured outputs + human-in-the-loop where outcomes are structured; reserve multi-step/multi-agent orchestration (Azure AI Foundry Agent Service, LangGraph / Microsoft Agent Framework) for genuinely open-ended tasks, keeping state-changing actions gated.
• Strengthen MLOps / LLMOps. Improve prompt/version management, model adaptation, CI/CD and the path from experiment to production; treat prompts and retrieval as versioned, tested production assets.
• Build for multi-tenancy. Default to per-product / per-tenant scoping of context, tools and actions; bake in observability (OpenTelemetry, Grafana, distributed tracing) and per-product cost/FinOps visibility.
• Help advance security, safety & governance. Inherit platform RBAC (Entra ID / AccessIT), defend against the OWASP LLM Top 10, keep AI usage auditable, and respect BASF / EU AI Act and data-residency requirements.