Lead LLM Engineer — Legal-Tech AI Agent (Paris)

The context

A leading French notary firm has entrusted us with designing an AI agent that assists notaries in drafting real estate deeds. The goal: reduce document generation time by 60% while ensuring regulatory compliance.

Your missions

• Architect the LLM agent (Claude Sonnet 4.5 + tool use) that consumes a document ontology (RDF/SPARQL) and generates notarial deeds.
• Set up RAG + knowledge graph for French legal context (Code civil, case law).
• Continuous evaluation: LangFuse evals, golden dataset, regression CI.
• Autonomous triggers: event detection, human escalation.

The profile

• 5+ years in production on LLM/agent projects (Claude, OpenAI, or similar frameworks).
• Strong in Python (strict typing, async, tests).
• Hands-on experience with tool use, RAG, and continuous evals.
• Knowledge of knowledge graphs (RDF, SPARQL, or similar) a plus.
• Interest in law/legal-tech: you're not afraid to read a Code civil.
• SEMANTIC observability in prod: you can detect a quality regression (continuous eval, versioned golden dataset, drift detection on new document formats), not just latency/cost monitoring.
• Advanced agentic patterns: graph state interruption/persistence (LangGraph & co) to handle long async tasks without blocking the agent or burning tokens; genuinely multi-tool, iterating agents.
• Lead posture: you bring the patterns and raise the bar, rather than executing a pre-made scope.