Senior Agentic AI Engineer

Project overview

Context: our client will rely on an agent platform on AWS to assist GPPO teams in processing international documents and cases. VITAL maintains business data, deterministic rules, and transaction control.

Mission: Design, develop, and industrialize the client's AI agents on the agent platform, integrating them with client services and human validation workflows.

Main responsibilities

Skills sought

Profile:

• Develop agents and their workflows in Python and LangGraph: orchestration, prompts, context, and tool calls.
• Implement documentary uses: classification, data extraction from invoices, multilingual summarization and translation, consistency checks.
• Connect agents to the client's MCP server and define tool contracts with the full-stack developer.
• Utilize documents and the graph knowledge base through client-exposed services and tools.
• Implement the reception and processing of VITAL events on the agent platform side, as well as workflow resumption.
• Develop synchronous copilot-type interactions and human validation steps.
• Build tests and evaluations: relevance, errors, robustness, latency, and cost.
• Instrument agents, protect sensitive data in traces, and contribute to YAML configurations and deployments with the platform team.
• Solid mastery of Python and concrete experience developing applications using LLMs.
• Practice with LangGraph, tool calls, structured outputs, and client-side MCP protocol.
• Experience with APIs, synchronous calls, and asynchronous event processing.
• Desired skills: Amazon Bedrock, AgentCore, AWS, observability, and CI/CD.
• Ability to evaluate AI results and implement safeguards.
• Sensitivity to multilingual document processing and health data confidentiality.
• Senior engineer focused on development and industrialization, capable of working with business experts, VITAL developers, and the agent platform team. Professional English desired. Experience in insurance or health is a plus.
• Scope: the shared infrastructure of the agent platform remains the responsibility of the platform team. The availability of a RAG service is not presumed.

Project details

• Start: flexible
• Remote: hybrid
• Location: Saint-Denis, France, FR
• Languages: French