Principal AI Engineer

MiniMed•Atlanta, GA

About The Position

MiniMed is building a lean, high-leverage AI & Data Science team. We are looking for a Principal AI Engineer to be our senior technical anchor — the person who builds AI capabilities hands-on and sets the standard the rest of the team builds to. This is a builder’s role first. You will take AI/ML models and LLM-powered agents from a fast “proof of life” prototype through to a production deployment that holds up in a regulated environment. Roughly a fifth of your time goes to the force-multiplier work: reviewing and quality-gating other engineers’ designs, and establishing the reusable patterns that keep the architecture and the hard-won judgment in-house. To be clear about what this is not: this is not a people-management role, and it is not a role where you hand a notebook to someone else to productionize. You own the capability until it is live and delivering value.

Requirements

Bachelor’s degree in Computer Science, Engineering, or a related technical field and 8+ years of of relevant experience or advanced degree with a minimum of 6+ years of relevant experience.

Nice To Haves

Demonstrated delivery of LLM-powered systems or ML models to production — not prototypes, evaluations, or internal demos alone.
Strong system design skills and end-to-end ownership from prototype through production deployment.
Experience technically leading or directing other engineers, including reviewing their work.
Strong stakeholder-facing communication — able to scope ambiguous problems with business partners and explain technical trade-offs (latency, cost, model risk) to non-technical leaders.
Hands-on with Databricks, LangGraph, LangSmith, vector databases, and managed RAG/retrieval.
Forward-deployed or embedded engineering experience — building AI systems from inside the business’s real data, workflows, and constraints.
Experience integrating AI or software systems into complex enterprise environments — APIs, identity/auth, systems of record (e.g., Salesforce, SAP), and data pipelines.
Comfort operating within a formal MLOps discipline, including a defined handoff to a model operations function.
Experience in a regulated industry (medical device, healthcare, financial services); familiarity with HIPAA, audit, and validation requirements.

Responsibilities

Prototype fast. Demonstrate “proof of life” for AI/ML models, agents, and tools against real MiniMed business problems, working from the business’s actual data and workflows rather than a sanitized sandbox.
Take it to production and drive adoption. Build, evaluate, and deploy on the MIA stack (Databricks, LangGraph/LangSmith, enterprise agent platforms), applying evaluation-driven development, guardrails, and deployment patterns so what ships is reliable, auditable, and maintainable. You own the capability from prototype through initial production release and its first monitoring cycle — measured by real adoption and business value, not just a stable deployment — then hand off to Model Operations for steady-state run.
Integrate into the enterprise environment. Wire agents and models into the surrounding systems — APIs, identity and access (SSO/SAML/OAuth), systems of record such as Salesforce and SAP, and enterprise data pipelines — so capabilities work against real, messy production systems rather than in isolation.
Set the technical standard. Establish reusable patterns for agent design, tool-calling, retrieval, and evaluation harnesses; review and quality-gate the work of other engineers on the team so the bar holds without a manager in the loop.
Feed learnings back to the platform. Turn what you learn in the field into improvements to the MIA platform, shared tooling, and reusable-pattern roadmap, so each deployment makes the next one faster.
Self-direct against outcomes. Partner with the Product Manager, AI & Data Science to choose what to build and when to stop, without needing the problem pre-decomposed for you.
Build for a regulated environment. Uphold the safety, privacy, and compliance requirements of a medical-device context, including auditability and human-in-the-loop where required.