ML Engineering Lead (LLM Ops)
At a glance
Senior AI Engineering role at Neko Health. Stockholm · full-time.
Pay not stated AI Engineering salary
aimlroles summary, based on the employer's posting.
What you'll do
- Set up production tracing for prompts, tool use, retrieval, response times and costs across LLM and agent workflows
- Build LLM evaluation with automated scoring and reviewer feedback, replacing ad hoc checks with a repeatable process
- Put a RAG or agent pipeline into production, with version control and safe rollout, testing and rollback
- Document how cost, latency and GPU capacity shape choices among API, Databricks and self-hosted serving
- Connect LLM operations with the MLOps team to keep clinical workflows reliable
What you bring
- You have independently led complex production MLOps work across tracking, training and monitoring
- You’re fluent in Python and core ML concepts, with experience shipping complete ML systems and platform initiatives
- You’ve built LLM or GenAI applications using prompting, RAG, agents or chains and orchestration frameworks
- You understand PyTorch, distributed systems, ML orchestration and key trade-offs in operating LLM applications
Who this fits
This role suits a senior engineer focused on LLM, GenAI and RAG operations who can take ownership of production systems. You’ll work across clinical, regulatory, firmware and hardware contexts, and integrate closely with the MLOps team. The role is hybrid in Stockholm.
From the employer
Mission
Neko is redefining what prevention means, from treating illness when it arrives, to sustaining health before it's ever at risk. Our mission: make data-driven, preventative care accessible to more people, before symptoms appear.
In a single, non-invasive visit under an hour, proprietary technology and direct clinical care combine to deliver personalised, actionable insights. It's a team that thinks in 10x, not 10%. Every role here plays a part in building a world where prevention is the norm, and where your work genuinely helps people live longer, healthier lives.
Role Purpose
This role owns the operational lifecycle of Neko's LLM, GenAI and RAG based systems: deployment, monitoring, evaluation and iteration. It builds the production grade platform that lets clinical ML and GenAI workflows run reliably on proprietary sensor and device data, inside a regulated medical device QMS, across Skin, Cardio and other use cases. The role sits within the ML Engineering Area and integrates tightly with the existing MLOps team rather than as a silo.
What You'll Deliver in the First 6-12 Months
● Stand up MLflow Tracing observability, prompts, tool calls, retrievals, latency and cost, live across production LLM and agent pipelines.
● Build an evaluation suite combining built-in and custom LLM judges and scorers, with a human-feedback loop via review apps, replacing today's ad hoc review.
● Ship at least one RAG or agentic pipeline to production with prompt and application versioning through the MLflow Prompt Registry and Unity Catalog, enabling safe rollout, A/B testing and rollback.
● Produce a documented cost, latency and GPU capacity framework for choosing serving strategy: third party API versus Databricks External Models versus self-hosted.
● Integrate LLM Ops tightly with the existing MLOps team so it operates as a natural extension of the platform rather than a separate track, protecting reliability on clinical workflows.
Minimum Qualifications
● Solid MLOps fundamentals across the full lifecycle, experiment tracking, training and monitoring, demonstrated through independent ownership of complex, production grade work.
● Fluent in Python and core ML concepts, with a track record of shipping end-to-end production ML systems and platformisation initiatives.
● Practical, hands-on experience building LLM or GenAI applications: prompt engineering, RAG, agents or chains, using frameworks such as LangChain, LangGraph, or comparable orchestration tools.
● Working knowledge of PyTorch, distributed systems and ML orchestration.
● Conceptual understanding of LLM-specific MLOps trade-offs: fine-tuning versus prompting versus RAG, vector databases, embedding models, and human-feedback loops for non-deterministic outputs. Hands-on fine-tuning ownership is not required, as execution sits with a separate track.
● Genuine, demonstrable motivation for LLM, GenAI and RAG work specifically, not generic MLOps, and the ability to navigate complex systems spanning the medical domain, regulation, firmware and hardware.
Preferred Qualifications
● Experience with agentic or AI-assisted coding workflows, provided you retain full ownership and understanding of the resulting output.
● Kubernetes and Terraform, useful for infrastructure as code or self-hosting fine-tuned or open-source models outside managed serving.
● Exposure to LLM evaluation and observability practices: tracing, LLM-as-judge, guardrails and safety scorers. Databricks MLflow 3 for GenAI experience is a strong plus.
● Comfort navigating a fast-moving tools and platform ecosystem, and distilling recommendations relevant to Neko's specific context.