Staff ML Engineer, AI Platform

Ambience Healthcare•San Francisco, CA

2d•$250,000 - $300,000•Hybrid

About The Position

Ambience Healthcare is the leading AI platform for documentation, coding, and clinical workflow, built to reduce administrative burden and protect revenue integrity at the point of care. Trusted by top health systems across North America, Ambience’s platform is live across outpatient, emergency, and inpatient settings, supporting more than 100 specialties with real-time, coding-aware documentation. The platform integrates directly with Epic, Oracle Cerner, athenahealth, and other major EHRs. Founded in 2020 by Mike Ng and Nikhil Buduma, Ambience is headquartered in San Francisco and backed by Oak HC/FT, Andreessen Horowitz (a16z), OpenAI Startup Fund, Kleiner Perkins, and other leading investors. Join us in the endeavor of accelerating the path to safe & useful clinical super intelligence by becoming part of our community of problem solvers, technologists, clinicians, and innovators. The Role: Ambience ships clinical AI to millions of patient encounters across the nation's largest health systems. How fast we improve that AI depends on the platform you'll own. You'll build evaluation and release gates that let teams ship confidently. Observability that surfaces quality issues before clinicians do. Debug tooling that makes reproducing regressions fast. The chart context retrieval layer that assembles patient history into model-ready inputs. The goal: teams iterate on quality in days, not weeks. Every improvement you make compounds across every product team, every quarter. Our engineering roles are hybrid in our SF office (3x/week).

Requirements

7+ years in software engineering, 3+ focused on ML infrastructure, platform engineering, or data systems
Staff-level scope: owned cross-cutting infrastructure, influenced technical direction across multiple teams
Strong backend fundamentals in Python, TypeScript, or similar
Built eval systems, data pipelines, or ML observability infrastructure
Comfortable on both the ML and Eng sides of MLOps
Track record of platform work that measurably accelerated other teams
In SF, 3x/week in-person

Responsibilities

Eval & Release Infrastructure — Automated graders and release gates that work across product pods. Unified eval dataset versioning and execution to replace fragmented workflows. Production quality monitoring with end-to-end tracing, shared metrics, and automated alerting.
Debug Tooling — Encounter replay that reconstructs exact inference inputs (retrieved chart context, packed prompts, model versions) so teams reproduce issues without digging through logs. Diff views comparing known-good runs to regressions.
Chart Context & Data Pipelines — The retrieval layer that pulls relevant patient history and assembles it into consistent model-ready inputs. Feedback loops that capture real-world usage and convert it into training signal. End-to-end latency instrumentation across every workflow step.
Preference Infrastructure — The system that enables clinician and site-specific behavior across specialties. Different clinics want different defaults, different phrasing, different workflows. You'll build the platform that supports customization at scale.
Model Serving — Performance and reliability layer for critical in-house models with clear SLOs, capacity planning, and regression alerts.