Software Development Engineer - II, FinTech - TFAW

Do you want to build machine learning and agentic-AI systems that analyze billions of dollars in transactions every day, and measure your impact in millions of dollars saved?

In Amazon's Finance Technology group, we build the detection systems (machine learning, and increasingly agentic AI) that defend Amazon against theft, fraud, abuse, and waste across the supply chain, retail catalog, and corporate procurement. We turn terabytes of transaction, vendor, and operational data into real-time risk decisions: scoring payments as they happen, surfacing anomalous behavior, and uncovering collusion networks that span thousands of entities.

We're now building the next generation of that platform around generative and agentic AI: multi-agent systems that discover fraud patterns directly from the raw data flowing through Amazon's businesses, working alongside domain experts to turn what they find into production detection logic. The result is a platform where new detection coverage comes not only from engineers, but from agents and the experts who know the fraud best, shipping in days instead of sprints.

You'll build the agent tools, ML pipelines, services, and data infrastructure behind all of this, working with generative AI, large language models, and agent frameworks alongside the anomaly-detection, classification, and graph-modeling techniques they build on. You'll partner directly with applied scientists, fellow engineers, and finance operations teams, and you'll see your ideas reach production in weeks, not years.

If you want to apply advanced ML and agentic AI at Amazon scale, on systems with a direct, measurable financial impact, we'd love to talk.

Key job responsibilities You will participate in the full software development lifecycle, from collaborating with customers and scientists on design, to building scalable, extensible systems that run in production. You will:

• Design and build ML pipelines that process terabytes of data and score billions of dollars in transactions.
• Build agentic AI applications: LLM-powered agents that discover risk patterns and generate detection logic, operating through well-defined tool interfaces with human-in-the-loop review.
• Build the knowledge and memory systems agents depend on: vector stores, knowledge graphs, the entity-risk layer for cross-entity collusion detection, and in-session and shared memory.
• Develop how we evaluate agent effectiveness in fraud detection, and turn analysts' ad-hoc investigation patterns into repeatable, templated workflows.
• Build the platform and tooling (governed data access, model deployment, and self-service rule authoring) that lets agents and non-engineers ship detection logic safely and fast.