Senior Machine Learning Engineer, Alexa-Conv A Modeling&Learning

Alexa AI is building the next generation of Alexa+, Amazon's LLM-powered conversational assistant, and its future is agentic: LLM systems that reason and act over dozens of chained inferences, coupled to real environments where their actions persist. Making these agents smarter, faster, and cheaper is as much a systems problem as a modeling problem - agent performance depends on the model, the harness, the evaluation infrastructure, and the serving stack co-designed together. We are looking for a Senior Machine Learning Engineer to build and own core systems in this agentic platform. You will take one of its foundational areas - agentic evaluation infrastructure, reinforcement learning training systems, self-learning pipelines, or agentic inference serving - and own it end to end: the design, the implementation, the operational bar, and the interfaces that scientists and partner teams build on. You will work directly with applied scientists, work backwards from committed product launches, and turn research prototypes into infrastructure that runs unattended at scale. The work is concrete. Agents are evaluated in sandboxed, recreatable environments at hundreds of concurrent trials, and every source of infrastructure noise you remove is a model decision the organization can trust. They are trained on long-horizon multi-turn trajectories where the rollout and learner engines have to agree token for token. They are served under latency budgets measured in hundreds of milliseconds. And they improve week over week only if the pipeline that turns production experience into training data actually holds. You will own a piece of that loop, make it reliable, and make it fast. This is a platform role with room to grow. The systems you own serve every Alexa agent program rather than a single product, and the engineer who makes them dependable becomes the person the organization routes its hardest cross-system problems to. Key job responsibilities Design, build, and operate major components of the agentic AI platform: evaluation harnesses, sandboxed environments and mocked resources, RL and post-training pipelines, self-learning data pipelines, or inference serving for agentic traffic Lead the design work in your area: write the design documents, drive them through review, and make the build-versus-adopt calls within your scope Own reliability and performance: instrument your systems, drive down the failure modes that make results untrustworthy (process management, resource contention, unreliable external calls), and report platform health in metrics rather than anecdotes Partner with applied scientists to turn research code into production infrastructure, and expose it through interfaces other teams can use without your involvement Scale what you build: hundreds of concurrent evaluation trials, long-context multi-turn training jobs, and large GPU formations on shared company infrastructure Raise the engineering bar through code and design reviews, operational excellence practices, and deep dives on cross-system problems Mentor engineers earlier in their careers and help set technical direction for your team A day in the life You might spend the morning making the evaluation platform reproducible under high concurrency, tracking down why scores drift when a hundred trials share a host, midday pairing with a scientist to get a long-context training job to converge identically across the rollout and learner engines, and the afternoon in a design review deciding how environment snapshots should be versioned and served to partner teams. You work daily with applied scientists and other engineers, and your systems are the reason their results are trustworthy and shippable.

Our organization owns the applied science and platform engineering for Alexa's agentic experiences. We operate at the intersection of large language models, reinforcement learning with verifiable rewards, agentic architectures, and large-scale distributed