LLM Engineer

Project overview

Mission freelance – Generative AI Engineer: RAG & Agents (M/F)

Client: large insurance company, Paris

Project details

ASAP | Duration: 12 months renewable

Frequency: 2 to 3 days on-site

according to profile

Experience: 3 years minimum in ML/AI, including 1 to 2 years on LLMs in production

Context

The AI department is developing business assistants and agents for claims management, underwriting, and customer support. Several RAGs are already in production, and the current challenge is twofold: improving their quality and moving to autonomous agents connected to the information system.

Missions

RAG

Agents

Quality and production

Technical Stack

Profile sought

Essential

Appreciated

• Design advanced RAG pipelines: parsing complex documents (PDFs, tables, scans), chunking strategies, hybrid search (vector + BM25), reranking
• Implement agentic RAG: query reformulation, routing between sources, multi-step search
• Reduce hallucinations and improve source citation reliability
• Develop business agents with tool calls: internal APIs, databases, document generation
• Design multi-agent architectures and workflows with human validation
• Expose IS tools and data via the MCP protocol
• Manage agent memory, state, and error recovery
• Build evaluation datasets and metrics: source fidelity, relevance, success rate of agent tasks
• Trace and debug agent executions
• Implement safeguards: prompt injections, personal data, agent scope of action
• Optimize latency and cost per query
• LLM: Azure OpenAI, Mistral, Claude
• Agent orchestration: LangGraph, MCP
• RAG: LlamaIndex, Docling, Azure AI Search, Qdrant, rerankers (Cohere, BGE)
• Evaluation and observability: Ragas, DeepEval, Langfuse
• Backend: Python, FastAPI, Pydantic
• Deployment: Docker, Kubernetes, GitLab CI
• At least one RAG in production with real users, and the ability to explain how its quality was measured and improved
• At least one agent with tool calls developed beyond the POC stage
• Solid Python backend skills (API, asynchronous, testing)
• GraphRAG or knowledge graphs
• Light fine-tuning (LoRA) or custom embeddings
• Insurance or banking sector
• Start: flexible
• Remote: hybrid
• Location: Paris, France, FR
• Languages: French