Staff PlatformOps Engineer (Emerging AI)

DAT•Seattle, WA

5d•Hybrid

About The Position

As a Staff AI Platform Engineer, you'll build and own the platform that every AI and machine learning workload at DAT runs on. Freight is an uncertain business, and the models we ship reduce that uncertainty: rate forecasts, load-to-truck matching, document extraction, fraud signals, and the agentic workflows our brokers and carriers use to move freight faster. None of that reaches a customer without a platform that makes training, serving, evaluating, and monitoring models routine instead of heroic. You'll set the architectural direction for that platform, from the model gateway and inference layer through feature and vector storage, evaluation harnesses, and production observability. This is a highly visible role for an engineer who wants their work multiplied across every AI team in the company.

Requirements

10+ years of experience in software engineering, including significant time as a senior or staff engineer owning production infrastructure that other engineering teams depend on.
4+ years building and operating machine learning or AI platforms, with models you've taken from notebook to production traffic and then kept healthy.
Strong development language skills ideally using TypeScript/ Node.js, Java, or Go, with the software engineering discipline to build libraries other teams adopt willingly.
Extensive AWS experience with production systems, ideally including EKS/Kubernetes, Lambda, Redpanda, MSK/Kafka, S3, and Secrets Manager, plus infrastructure-as-code with Pulumi, Terraform or similar.
Hands-on experience serving models at scale: containerized inference, GPU scheduling, autoscaling, batching, and the tradeoffs between real-time, streaming, and batch scoring.
Practical experience with LLM systems in production, including retrieval-augmented generation, vector databases, prompt and context management, tool-calling or agent frameworks, and managed model APIs such as Amazon Bedrock.
Proven experience building evaluation and monitoring for models, not just services: offline eval sets, online metrics, drift and data quality checks, and a clear definition of what "regression" means for a model.
Strong background in event-driven and distributed systems using Redpanda, AWS MSK or Kafka
Fluency with observability and on-call health, designing dashboards, APM traces, logs, and alerts, defining SLOs, and using these tools to drive down incident frequency and MTTR.
Experience establishing and raising engineering standards for a team: PR and testing guidelines, deployment validation checklists, model release criteria, and post-incident review practices.
Demonstrated technical leadership: leading cross-team designs in ambiguous problem spaces, writing clear system design documents and operations runbooks, and driving alignment across Engineering, Product, and Operations.
Proven mentoring track record, especially helping engineers and data scientists grow in system design, observability, and operational excellence.
Excellent communication skills, with the ability to explain trade-offs and system behavior clearly to engineers, product managers, and non-technical stakeholders.

Nice To Haves

Experience with model fine-tuning, distillation, or parameter-efficient training, and a clear point of view on when it beats prompting.
Background in marketplace, pricing, forecasting, or document extraction problems.
Freight, logistics, or supply chain domain experience.

Responsibilities

Design, build, and operate the shared services that engineers and business users use to ship models: a model gateway for LLM access, inference endpoints for real-time and batch scoring, feature storage, vector search, and a common SDK.
Lead architecture for large-scale AI systems, write the design documents, and drive alignment across Product, and Engineering and Business users on how models get built and shipped at DAT.
Architect and run scalable, reliable AI infrastructure on AWS, including Bedrock, SageMaker, EKS, Redpanda, MSK (Kafka), Lambda, and S3, all defined in Pulumi.
Build the offline and online evaluation systems that tell us whether a model or prompt change is an improvement, including regression suites, LLM-as-judge pipelines, A/B and shadow testing, and drift detection.
Own inference cost and latency as first-class metrics. Right-size GPU and serverless capacity, tune batching, caching, and quantization, and give teams clear visibility into what their workloads cost.
Implement guardrails, prompt and output logging, PII handling, access controls, and model and dataset lineage so AI systems meet our security and customer data commitments.
Drive the adoption of modern software development practices across AI work, including automated testing, code reviews, CI/CD pipelines, and infrastructure-as-code.
Mentor engineers and data scientists on production ML and distributed systems, and raise the operational bar of every team that builds on the platform.
Lead the response and resolution for complex production incidents involving AI services, perform root cause analysis, and implement preventative measures.