Software Development Engineer, AWS Marketing, Data Science & Engineering (D:SE)
Posted Updated
Within AWS Marketing, the Data Science and Engineering (D:SE) team is the central data, science, and intelligence organization. We own the data foundation, the measurement and causal economics frameworks, the scoring and intelligence models, and the agentic AI systems that power marketing decisions across AWS.
We exist because marketing at AWS scale requires more than dashboards and reports. It requires a unified data layer rich enough to train models on, composable enough for AI agents to reason over, and rigorous enough to hold investment decisions accountable. We build that layer and the intelligence products on top of it.
We're looking for a Software Development Engineer to help us build and scale our next-generation MLOps and agentic AI systems. You'll work with a serverless, AWS-native stack:
- Agentic AI - AWS Bedrock and Bedrock AgentCore (runtime, memory, tool gateway), Bedrock Guardrails, ReAct-style agent loops with model routing, and Model Context Protocol (MCP) servers that expose our data and retrieval tools to agents
- ML platform - AWS SageMaker for training pipelines, Feature Store, and real-time inference endpoints; embedding pipelines; hybrid retrieval combining dense vector similarity and BM25; reranking; and LLM-as-judge evaluation harnesses
- Compute and orchestration - AWS Lambda, AWS Step Functions, AWS EventBridge, and Amazon Managed Workflows for Apache Airflow (MWAA)
- Data and serving - AWS Redshift (via the Redshift Data API), AWS S3, AWS DynamoDB, and AWS OpenSearch Service
- Infrastructure and operations - AWS CDK for all infrastructure, so every environment is reproducible and code-reviewed rather than clicked into a console; AWS CloudWatch metrics, alarms, and dashboards; CloudWatch RUM; SNS alerting; SQS dead-letter queues; and canary deployments promoted through beta, gamma, and production
Key job responsibilities
- Design, build, and operate production agentic AI systems - agent runtimes, MCP tool servers, retrieval pipelines, guardrails, and the evaluation harnesses that keep them honest
- Own the retrieval and ranking quality loop end to end: hybrid semantic and keyword retrieval, graded relevance signals, reranking on live in-session data, and offline plus LLM-as-judge evaluation before parameters become contract
- Productionize machine learning models with your applied science partners - training pipelines, feature engineering into SageMaker Feature Store, real-time inference behind low-latency APIs, and the deployment gates that make model rollout safe
- Build and scale serverless data pipelines over datasets in the billions of rows, including ingestion connectors, transformation orchestration, and validated migrations with parallel-run and rollback strategies
- Raise the operational bar on what you own: instrumentation, actionable alarms tuned against real SLOs, data-quality and freshness observability, runbooks, load and game-day testing, and participation in an on-call rotation
- Write the design documents, drive the code reviews, and make the tradeoff calls - you'll own systems, not tickets
- Work AI-natively. Our team uses agentic development tooling daily, and we expect you to extend it as well as use it
A day in the life
You might start by triaging an alarm on an agent's tool-call latency, then pair with an applied scientist on why a reranking signal isn't lifting conversion, then review a teammate's CDK change that adds a freshness alarm to an indexing pipeline. Afternoons tend toward deeper work: a design doc for replacing a mocked data source with a live pipeline, or an evaluation run that decides whether a model change ships. You'll spend meaningful time with product and data engineering partners, because most of our interesting problems are ambiguous before they're technical.
About the team
You'll join a tight, high-impact team of software engineers, ML engineers, data engineers, applied scientists, and product managers solving problems at the intersection of marketing analytics, data science enablement, and platform engineering. We're small enough that your work is visible and unambiguously yours, and we ship fast - recent agentic AI products have gone from concept to production in a handful of sprints. You'll experience a culture that values ownership, cross-functional collaboration, and data-driven decision making.
We exist because marketing at AWS scale requires more than dashboards and reports. It requires a unified data layer rich enough to train models on, composable enough for AI agents to reason over, and rigorous enough to hold investment decisions accountable. We build that layer and the intelligence products on top of it.
We're looking for a Software Development Engineer to help us build and scale our next-generation MLOps and agentic AI systems. You'll work with a serverless, AWS-native stack:
- Agentic AI - AWS Bedrock and Bedrock AgentCore (runtime, memory, tool gateway), Bedrock Guardrails, ReAct-style agent loops with model routing, and Model Context Protocol (MCP) servers that expose our data and retrieval tools to agents
- ML platform - AWS SageMaker for training pipelines, Feature Store, and real-time inference endpoints; embedding pipelines; hybrid retrieval combining dense vector similarity and BM25; reranking; and LLM-as-judge evaluation harnesses
- Compute and orchestration - AWS Lambda, AWS Step Functions, AWS EventBridge, and Amazon Managed Workflows for Apache Airflow (MWAA)
- Data and serving - AWS Redshift (via the Redshift Data API), AWS S3, AWS DynamoDB, and AWS OpenSearch Service
- Infrastructure and operations - AWS CDK for all infrastructure, so every environment is reproducible and code-reviewed rather than clicked into a console; AWS CloudWatch metrics, alarms, and dashboards; CloudWatch RUM; SNS alerting; SQS dead-letter queues; and canary deployments promoted through beta, gamma, and production
Key job responsibilities
- Design, build, and operate production agentic AI systems - agent runtimes, MCP tool servers, retrieval pipelines, guardrails, and the evaluation harnesses that keep them honest
- Own the retrieval and ranking quality loop end to end: hybrid semantic and keyword retrieval, graded relevance signals, reranking on live in-session data, and offline plus LLM-as-judge evaluation before parameters become contract
- Productionize machine learning models with your applied science partners - training pipelines, feature engineering into SageMaker Feature Store, real-time inference behind low-latency APIs, and the deployment gates that make model rollout safe
- Build and scale serverless data pipelines over datasets in the billions of rows, including ingestion connectors, transformation orchestration, and validated migrations with parallel-run and rollback strategies
- Raise the operational bar on what you own: instrumentation, actionable alarms tuned against real SLOs, data-quality and freshness observability, runbooks, load and game-day testing, and participation in an on-call rotation
- Write the design documents, drive the code reviews, and make the tradeoff calls - you'll own systems, not tickets
- Work AI-natively. Our team uses agentic development tooling daily, and we expect you to extend it as well as use it
A day in the life
You might start by triaging an alarm on an agent's tool-call latency, then pair with an applied scientist on why a reranking signal isn't lifting conversion, then review a teammate's CDK change that adds a freshness alarm to an indexing pipeline. Afternoons tend toward deeper work: a design doc for replacing a mocked data source with a live pipeline, or an evaluation run that decides whether a model change ships. You'll spend meaningful time with product and data engineering partners, because most of our interesting problems are ambiguous before they're technical.
About the team
You'll join a tight, high-impact team of software engineers, ML engineers, data engineers, applied scientists, and product managers solving problems at the intersection of marketing analytics, data science enablement, and platform engineering. We're small enough that your work is visible and unambiguously yours, and we ship fast - recent agentic AI products have gone from concept to production in a handful of sprints. You'll experience a culture that values ownership, cross-functional collaboration, and data-driven decision making.