AI Native Software Engineer

Accenture France; Paris; Sur place; EUR 90 000 - 140 000; Plein temps

Résumé du poste

Accenture France is seeking an AI Engineer (Agentic/Applied) to design, build, and deploy production-grade agentic AI systems across the enterprise tech stack, working directly with client engineering teams. You will lead design sessions, implement RAG pipelines, and create reusable patterns to scale across engagements.

This role requires hands-on production experience, strong Python/Java skills, and the ability to integrate multiple LLM providers while ensuring observability and cost efficiency

Qualifications

• 2 to 5 years of software engineering experience in production environments.
• Minimum 1 year hands-on experience deploying agentic AI solutions in production.
• Experience with agentic orchestration frameworks in production depth.
• Direct experience calling LLM APIs (OpenAI, Anthropic, Vertex AI) in production code.
• RAG pipeline ownership: embeddings, chunking, vector databases, and context engineering.
• LLMOps fundamentals: eval harness design, prompt versioning, and production observability.
• Cloud-native engineering maturity: Kubernetes, Docker, microservices, CI/CD, and IaC.
• Strong Python; Java or equivalent backend language; production debugging and observability.
• Quality of experience valued over years; shipped production agentic systems preferred.

Responsabilités

• Design and build production-grade agentic systems end-to-end: multi-agent orchestration, RAG pipelines, policy-based routing, tool invocation, memory management, and lifecycle observability.
• Build and own RAG pipelines: embeddings, chunking strategy, vector search, context window engineering and tuning against real quality targets.
• Integrate and abstract across multiple LLM providers — OpenAI, Anthropic, Vertex AI, and open-source models — with fallback routing, token, cost, and latency management.
• Implement LLMOps in production: eval harnesses with real quality metrics, prompt versioning, observability tooling, cost and safety monitoring.
• Embed directly with client engineering teams to design, prototype, and deploy agentic solutions — workshops, proofs of concept, code-with sessions, and architecture walkthroughs.
• Build reusable patterns, accelerators, and playbooks that scale beyond the individual client engagement.
• Define and use metrics to measure agent accuracy, latency, safety, and cost-effectiveness; present findings to client stakeholders.

Connaissances

Python

Java

Production debugging

LLM integration

Outils

LangGraph

CrewAI

AutoGen

LangSmith

Braintrust

Kubernetes

Docker

Terraform

Helm

Description du poste

You build the systems that actually make AI work in enterprise environments, not demos, not prototypes that stall after a pilot, but production agentic architectures running inside real client organizations. The difference between an AI Engineer and what we are looking for is straightforward: you have shipped a multi-agent system in production, you have owned the eval harness, and you know what happens when your agent fails at 2am because you have lived it.

As an AI Engineer (Agentic/Applied), you will design, build, and deploy production-grade agentic AI systems across the full enterprise technology stack. You will work directly with client engineering teams, lead technical design sessions, and build reusable patterns and accelerators that scale beyond individual engagements.

This role sits at the heart of the AI engineering talent market — demand is growing faster than supply and will continue to do so. We offer what no single product company can: breadth across every industry, every enterprise technology stack, and every level of organizational complexity, combined with vendor fellowship access inside Anthropic, OpenAI, Microsoft, and Google engineering teams and a direct pathway to the Forward Deployed Engineer programme.

Key Responsibilities

• Design and build production-grade agentic systems end-to-end: multi-agent orchestration, RAG pipelines, policy-based routing, tool invocation, memory management, and lifecycle observability
  
• Build and own RAG pipelines: embeddings, chunking strategy, vector search, context window engineering and tuning against real quality targets
  
• Integrate and abstract across multiple LLM providers — OpenAI, Anthropic, Vertex AI, and open-source models — with fallback routing, token, cost, and latency management
  
• Implement LLMOps in production: eval harnesses with real quality metrics, prompt versioning, observability tooling (LangSmith, Braintrust, or equivalent), cost and safety monitoring
  
• Embed directly with client engineering teams to design, prototype, and deploy agentic solutions — workshops, proofs of concept, code-with sessions, and architecture walkthroughs
  
• Build reusable patterns, accelerators, and playbooks that scale beyond the individual client engagement and enable the next one to start faster
  
• Define and use metrics to measure agent accuracy, latency, safety, and cost-effectiveness; present findings and recommendations to client stakeholders in business terms
  
• 2 to 5 years of software engineering experience in production environments
  
• Minimum 1 year of hands-on experience designing and deploying agentic AI solutions in a production environment — non-negotiable
  
• Demonstrated experience with agentic orchestration frameworks: LangGraph, CrewAI, AutoGen, or equivalent — at production depth, not tutorial level
  
• Direct experience calling LLM APIs (OpenAI, Anthropic, Vertex AI) in production code: provider abstraction, token management, latency and cost tradeoffs
  
• RAG pipeline ownership: embeddings, chunking strategy, vector databases, and context engineering
  
• LLMOps fundamentals: eval harness design, prompt versioning, and production observability
  
• Cloud-native engineering maturity: Kubernetes, Docker, microservices, serverless, CI/CD, and IaC (Terraform or Helm)
  
• Strong Python; Java or equivalent backend language acceptable; production debugging and observability experience
  
• Quality of experience is weighted over years, a candidate who has shipped three production agentic systems in four years is preferred over a generalist with passive AI exposure