Lead ML/AI Engineer: Build Production-Grade AI for Health, Paris

Requirements

Advanced Python proficiency; solid experience with ML frameworks (TensorFlow, PyTorch, scikit-learn) , Strong SQL skills: complex queries, performance tuning, data modeling basics , Generative AI expertise: LLM APIs (OpenAI, Claude,…), LangChain/LlamaIndex , MLOps experience: CI/CD pipelines, model monitoring, deployment at scale, Cloud platform experience (AWS/GCP/Azure) and managed ML services (SageMaker, Bedrock, Vertex AI) , Product mindset with a bias for measurable impact and ROI , Clear communication with non-technical partners; ability to write crisp documentation (ex: diving / specifications / exploration / problem statement /…), Strong ownership and autonomy; pragmatic problem-solving approach , Collaborative spirit with Product, Design, and Engineering teams; embraces feedback culture What the job involves

As the first Senior ML/AI Engineer at Hublo, you will help us build our AI and Machine Learning capabilities from the ground up , We are at the beginning of this journey, and your primary mandate is to identify which problems are truly worth solving with AI and ML , As part of the Data Platform team, you will explore, prototype, and ship intelligent features powered by ML and LLMs that deliver measurable value to healthcare professionals, You'll work closely with Product to identify high-impact AI opportunities, then partner with both Engineering and Data teams to build, deploy, and maintain production solutions , This role directly addresses a core challenge: reducing administrative friction for caregivers and health managers so they can spend more time on patient care but only where AI proves to be the right solution, Own end-to-end delivery of AI/ML features, from problem framing to production deployment and iteration , Cover the full ML/AI scope: classical ML (recommendations, predictions, optimization) and LLM-based features (assistants, document understanding, search) , Run disciplined POCs with clear success metrics, baseline comparisons, and go/no-go criteria defined upfront with Product, Make pragmatic technical decisions on modeling approaches, data requirements, evaluation methods, and build vs buy trade-offs , Kill what doesn't work: document learnings from failed experiments and redirect resources quickly , Ensure production quality: latency, reliability, observability, security, and graceful degradation when models fail, Set initial standards for ML/LLM engineering: experiment tracking, prompt/model versioning, evaluation harnesses, and documentation templates , Build reusable components where it unblocks velocity: shared datasets, template pipelines, monitoring dashboards, keep it pragmatic , Lay the groundwork for MLOps/LLMOps: CI/CD for models, A/B testing infrastructure, basic drift/quality monitoring, Document "how we do AI at Hublo": evaluation rules, production checklists, and safety guidelines , Work closely with Product to identify high-impact AI use cases, shape scope based on feasibility, and align on success metrics , Partner with Engineering on integration, performance requirements, and operational reliability, own the AI/ML part, collaborate on the rest, Communicate clearly on uncertainty: especially for LLM limitations, expected quality, and trade-offs, set realistic expectations early , Lead and animate the AI Community of Practice: drive discussions, share patterns and learnings, ensure it stays active and valuable for the company , Grow AI literacy with short, practical sessions: what works when, how to evaluate AI outputs, common pitfalls to avoid, Share openly: write postmortems (including failures), document technical decisions, and make your work reusable , Encourage evidence over hype: measurable outcomes, honest limitations, and realistic timelines, set the tone for how Hublo builds AI , Uplift on key user outcomes from AI features (e.g., +X% reduced task duration,−Y% time-to-action), Feature adoption and retention for AI-powered workflows (weekly active users, repeat usage) , Quality gains: recommendation precision/recall, summarization quality scores, CSAT on AI features , Model and service latency and availability within agreed SLAs , Drift and incident rate kept below threshold; time-to-recovery after issues, Time from prototype to production; cadence of meaningful iterations per quarter , Experiment throughput with clear learnings (A/B tests, offline evaluations→ shipped features)

#J-18808-Ljbffr