Frontier Agentic AI Engineer

About the role

Santander AI Lab (CDAIO) is looking for a Frontier Agentic AI Engineer based out of Madrid, Spain. The AI Lab acts as the applied innovation engine of one of the world’s largest banks, building prototypes with partners such as Anthropic, AWS and others, and scaling them for real banking impact.

Key responsibilities

• Design, build and deploy production‑grade agentic AI systems with multi‑agent orchestration, real memory, planning and error recovery.
• Develop and fine‑tune small‑ and medium‑size language models for regulated banking use cases, including custom evaluation frameworks.
• Architect and implement Model Context Protocol (MCP) servers, A2A protocols and federated API layers to enable AI agents across the group’s multi‑country infrastructure.
• Prototype new ideas in two‑week sprints, delivering functional demos and concise one‑page summaries.
• Collaborate with researchers, data scientists and business stakeholders to translate technical concepts into tangible bank value.

Required profile

• 4–8 years of software or AI engineering experience, with at least 2 years building production LLM‑powered systems.
• Proven experience designing multi‑agent AI systems with memory management, stateful orchestration and graceful failure recovery.
• Hands‑on experience fine‑tuning language models (SFT, LoRA, RLHF) and creating evaluation pipelines.
• Bachelor’s or Master’s degree in Computer Science, Software Engineering, Mathematics, Physics or a related field.
• Professional proficiency in English and Spanish.

Required skills

• Advanced Python programming.
• Agentic AI frameworks such as LangGraph, AutoGen, CrewAI or Semantic Kernel.
• LLM APIs (Anthropic Claude, OpenAI, Llama, Mistral) and understanding of cost/latency trade‑offs.
• Cloud services on AWS (Bedrock, Lambda, SageMaker, S3).
• FastAPI for production‑grade API development.
• DevOps basics: Docker, Git, CI/CD pipelines.
• Model evaluation tools (LLM‑as‑judge, RAGAS, Promptfoo, Langfuse).
• Design and implementation of MCP servers (preferred).