Software Development Engineer, Benefits Technology

Amazon’s healthcare and benefit programs are among the largest investments the company makes in its people, and BXT is building one of the most ambitious AI-native platforms anywhere at Amazon to protect the value of those programs for more than a million employees and their families. This is not AI bolted onto an old process. It is a platform architected from the ground up around large language models and autonomous agents that reason over healthcare claims at a scale and depth no rules engine has ever matched — making sure the dollars Amazon spends on employee health reach the care employees actually need. Across a population this size, a small share of medical claims drive an outsized share of program cost. In partnership with Amazon’s health plan, we are building agentic systems that read live claims as they flow, reason about which ones merit a closer look, assemble the evidence, and route them for review before they are paid — so the program spends more on care and less on avoidable waste. The patterns worth finding never hold still, so the platform is built to learn: models retrain, agents improve from every outcome, and the system gets sharper every week. This is a big-data and AI-forward engineering role where the work has real consequences and the technology is genuinely frontier. You will build agentic systems that read live claims, reason about them with large language models, assemble their own evidence, and produce results a person can trust and act on. Correctness, latency, and availability are first-order requirements because these results feed decisions about benefit programs — and precision matters more than volume, because a wrong result costs an employee time and confidence in their benefits. Key job responsibilities • Design, build, and operate the AWS services that turn live claims data into signals Amazon can act on - the real-time pipelines, data stores, and scoring services - owning their correctness, latency, and availability alongside their results. • Build generative-AI in production, not beside it: LLM-backed evaluation, multi-agent workflows where specialized agents reason from different perspectives and assemble the evidence, retrieval over claims and reference data, and the test harnesses that keep model behavior predictable. • Build the models and the machinery around them - feature pipelines, scoring, threshold management, and the retraining path that keeps them sharp as patterns move. Routine volume resolves automatically; everything else routes to a person with the context they need. • Build the review surfaces a person works in: what the platform found, why it surfaced, the supporting data behind it, and a durable, auditable record of what was decided. • Design and operate the real-time data exchanges with Amazon’s health-plan partner, working directly with their engineers to solve entity resolution at scale - probabilistic matching, blocking strategies, and merge/split policy for records that share no common identifier. • Raise the engineering bar through code reviews, technical design, and sharing best practices.