Production-Grade LLM Systems Engineer

Leonar; Paris; Sur place; EUR 90 000 - 150 000; Plein temps

Résumé du poste

Leonar is building an AI coworker for legal and professional services, automating emails, document analysis, and complex workflows. The role focuses on owning and scaling a production AI system relied on daily, with direct collaboration with product and founders.

Your work will turn vague user problems into structured AI systems with measurable performance that improve weekly, addressing hallucinations and reliability in real-world workflows.

Qualifications

• Experience building production-grade LLM systems and architectures.
• Ability to design and implement evaluation frameworks and metrics.
• Strong debugging skills for complex model and system failures.
• Proven track record of shipping fast iteration loops from production data.
• Good technical judgment and decision-making under uncertainty.

Responsabilités

• Design and evolve the LLM/agent architecture for production use.
• Own output quality across key use cases (emails, document analysis, etc.).
• Build evaluation systems (datasets, metrics, regression detection).
• Drive fast iteration loops using production data and feedback.
• Improve retrieval, reasoning, and tool usage.
• Ensure production reliability (latency, failure modes, fallbacks).
• Collaborate with product and founders on what to build and why.

Connaissances

Shipping Real LLM Systems

Evaluation-Driven Development

Debugging Complex Failures

Speed of Iteration

Strong Technical Judgment

Outils

Python (FastAPI)

Postgres

GCP

LangGraph

LangChain

PostHog

Langfuse

Azure OpenAI

Description du poste

Leonar is building an AI coworker for legal and professional services, automating emails, document analysis, and complex workflows. The role focuses on owning and scaling a production AI system relied on daily, with direct collaboration with product and founders.

Your work will turn vague user problems into structured AI systems with measurable performance that improve weekly, addressing hallucinations and reliability in real-world workflows.