Generative & Agentic AI Architect

Project overview

Context

Seniority 9 years and more

Reporting to the Architecture and Private Cloud team, responsible for supporting projects around containerized applications in terms of architecture, design, and CI/CD, for the Group's various business units and for the benefit of internal employees and clients.

The AI Architect is responsible for the design and consistency of the Group's generative and agentic AI solution architectures. They will support their industrialization on private and public cloud, guide technological choices (models, frameworks, platforms), ensure security, compliance, and cost control, and disseminate best practices within the teams.

MISSIONS:

In particular, the position contributes to the implementation of the Group's agentic AI plan and facilitates the integration of AI into the development chain. As such, missions include:

Profile sought

Specific Expertise:

o Generative and Agentic AI: mastery of LLMs (GPT, Claude, Gemini, Mistral, Llama), prompt engineering, RAG architectures, and agentic frameworks (LangChain, LangGraph, LlamaIndex, CrewAI, AutoGen, Google ADK). Knowledge of MCP and A2A protocols.

o AI Platforms: experience with Vertex AI, Azure OpenAI, and/or Amazon Bedrock. Experience serving open-source models on private cloud (OpenShift AI, vLLM, KServe, Ollama) and model gateways (LiteLLM). Knowledge of GPU constraints.

o Data and Vector Databases: knowledge of pgvector, Milvus, Qdrant, and OpenSearch / Elasticsearch. Mastery of chunking, embeddings, hybrid search, and reranking. Good command of SQL and relational databases, particularly PostgreSQL.

o LLMOps and Quality: experience in model and agent evaluation (Ragas, DeepEval), observability and traceability (Langfuse, OpenTelemetry, MLflow), guardrails, prompt management, and cost monitoring (FinOps).

o Containers and DevOps: excellent command of Docker, Kubernetes, and OpenShift. Experience with CI/CD pipelines (Git, Jira, Jenkins) and GitOps approaches with ArgoCD. Advanced use of AI code assistants (Claude Code, GitHub Copilot).

o Architecture, Security, and Governance: knowledge of private cloud infrastructure components (network, storage, reverse proxy, TLS certificates), AI application security (OWASP Top 10 for LLM, secrets management, RBAC), AI Act, and responsible AI challenges.

• AI Solutions Architecture: design target architectures for generative and agentic AI use cases (assistants, RAG, autonomous agents, multi-agents); define reference patterns and reusable components.
• Watch and Community: ensure constant technological watch on the AI ecosystem; be proactive in proposing market solutions; contribute to fostering the Group's agentic AI community.
• AI Platform: contribute to the design and evolution of the Group's AI platform (LLM serving, model gateway, vector databases, agent orchestration) on OpenShift and trusted public cloud.
• Security, Compliance, and Governance: integrate security requirements (prompt injection, data leakage, agent authorization management), regulatory requirements (AI Act, DORA, GDPR), and responsible AI principles; participate in security validation committees.
• Quality and Performance: define model and agent evaluation strategies (relevance, hallucinations, latency); implement LLMOps observability and FinOps monitoring of model consumption.
• Writing and Collaboration: write architecture documents and technical specifications; work in close collaboration with IT and business teams; participate in architecture reviews.
• Industrialization: technically lead the transition of POCs and pilots to production, in conjunction with infrastructure, security, and production teams; ensure the stability, run, and evolution of deployed solutions.
• Design and prototype agents and agentic workflows, and participate in defining the modular architecture of the AI platform.
• Implement the exposure of SI tools and data to agents via the MCP (Model Context Protocol) and Group APIs, in a secure and governed manner.
• Industrialize the deployment of models and agents using a GitOps approach (ArgoCD, OpenShift GitOps) and automate LLMOps pipelines (evaluation, prompt versioning, monitoring).
• Act as the team's AI referent; mentor and support developers in adopting AI and code assistants.
• Take charge of support and project assistance.
• At least 9 years of experience in software development and architecture, including at least 3 years on AI or generative AI projects. Excellent command of Python (FastAPI) and good knowledge of Java (Quarkus, LangChain4j).

Project details

• Start: ASAP
• Duration: 18 mois
• Remote: hybrid
• Location: Paris, Île-de-France, FR
• Languages: French