Staff Machine Learning Engineer, Platform (MLOPS)

Credit Acceptance Corporation•San Francisco, CA

2d•$184,943 - $258,920•Hybrid

About The Position

Credit Acceptance is hiring a Staff Machine Learning Engineer to build and operate its machine learning and generative AI platform. The role owns the end-to-end deployment path for ML and GenAI models, including training and inference pipelines, model registry and versioning, serving endpoints, and controlled promotion across development, QA, and production. It also owns production runtime health, agent operations, evaluation and observability, platform cost management, and reusable tools and patterns that help product teams adopt the standard platform path. The engineer partners with Cloud Engineering, Data Engineering, Security, and SRE to integrate the ML platform with enterprise governance, identity, and observability. The role responds to AI-specific production incidents, maintains architecture documentation and system diagrams, and mentors engineers and interns. This position follows a hybrid work model, with two work-from-home days and three in-office days each week. Candidates must live in or be willing to relocate to the greater Southfield, Michigan area; the greater Bellevue, Washington area; or the San Francisco Bay Area, California.

Requirements

Bachelor's degree in Computer Science, Engineering, Statistics, or a relevant technical field and at least 7 years of relevant experience, or a master's degree in one of those fields and at least 5 years of relevant experience.
At least 5 years building and operating production ML or AI systems, with direct ownership of at least two of: training or inference pipelines, model serving infrastructure, model registry and versioning, or production monitoring and alerting.
Demonstrated ownership of a production ML or AI service through deployment, monitoring, on-call response, failure diagnosis, and delivery of a fix.
Strong Python and SQL skills, with production-quality practices including version control, testing, code review, and CI/CD applied to ML workloads.
Hands-on production experience with a cloud ML platform.
Working knowledge of operational differences between LLM/GenAI workloads and traditional ML, including token cost and latency behavior, caching and batching, and non-deterministic output.
Experience operating LLM or agent applications behind a gateway or proxy layer, and ability to develop best practices for model routing and fallback, credential and key management, rate limiting, and budget enforcement.
Working knowledge of agent tool-calling architecture, including the Model Context Protocol, MCP server functions, tool scoping and authorization, and gateway-brokered tool access.
Experience with containerization and infrastructure as code.
Ability to communicate technical and non-technical trade-offs clearly in writing to engineering and non-engineering audiences.
Must reside in or be willing to relocate to the greater Southfield, Michigan area; the greater Bellevue, Washington area; or the San Francisco Bay Area, California.
Required degrees must be from institutions accredited by the Council for Higher Education Accreditation or an equivalent accrediting body.

Nice To Haves

AWS and Databricks production experience, including model serving, job orchestration, and a model registry or experiment tracking system such as MLflow.
Production experience with GPU-backed model serving, including autoscaling under concurrent load, cold-start management, and cost control.
Experience with OpenTelemetry and an enterprise observability platform such as Dynatrace, including instrumentation of AI workloads.
Experience with model-serving efficiency techniques such as quantization, parameter-efficient fine-tuning, distillation, or inference optimization.
Experience with data governance and access control on a lakehouse platform, such as Databricks Unity Catalog.
Experience operating AI systems in a regulated industry, particularly financial services, including auditability, retention, and access-control requirements.
Experience with production agentic or multi-step AI systems, including tracing and debugging multi-turn behavior.
Hands-on experience with a managed agent or tool gateway, such as AWS Bedrock AgentCore Gateway, and with running MCP servers in a governed environment.
Familiarity with agent interoperability and identity standards, including A2A-style delegation, agent discovery and capability advertisement, and workload identity for non-human actors.

Responsibilities

Own the end-to-end deployment path for ML and GenAI models, including training and inference pipelines, model registry and versioning, serving endpoints, and controlled promotion across development, QA, and production.
Own production model runtime health, including monitoring, alerting, drift and quality-regression detection, latency and throughput objectives, capacity and autoscaling behavior, incident response, root-cause analysis, and corrective actions.
Operate the agent runtime layer by routing production agents through the enterprise AI Gateway and MCP Gateway, migrating existing agents to that path, and keeping tool surfaces scoped, versioned, and least-privilege.
Own production agent evaluation through online scoring and behavioral monitoring, baseline-based regression and drift detection, sampling and judge pipelines, and release gates.
Build and maintain the observability and evaluation substrate, including trace and telemetry capture for multi-turn and multi-step agent traces, logging standards, evaluation pipeline plumbing, and underlying data contracts.
Measure and manage platform unit economics, including cost per inference, document, and interaction, and produce cost analyses to inform build-versus-buy and hosting decisions.
Deliver reusable pipeline templates, deployment patterns, reference implementations, and internal tooling that enable product teams to adopt standard platform practices.
Partner with Cloud Engineering, Data Engineering, Security, and SRE to integrate the ML platform with enterprise governance, identity, and observability.
Respond to AI-specific production incidents, including prompt injection attempts, rogue-agent cost spikes, data-classification exposure through tool calls, delegation abuse, and model endpoint failures; drive incidents through corrective action.
Maintain accurate ML platform architecture documentation and system diagrams for design reviews.
Mentor engineers and interns in production ML practice and raise engineering standards through design and code review.