Description
Our generative AI products are only as good as the data behind them. This role owns that data layer from end to end: the pipelines that bring data in, the transformations that shape it, and the way it reaches retrieval systems, agents, and analytics. The work runs on AWS, and the aim is a single governed source that every consumer can rely on.
We want someone who has already built data pipelines for AI systems, not only for reporting. Preparing data for an LLM or an agent brings its own work around chunking, embeddings, indexing, and keeping content current, and you have done it before. The team is small and spans several languages, so you will own your pipelines and help set the standards the rest of us follow.
What you will do
- Build and run the batch and streaming pipelines that move data from source systems into the lake and through to the warehouse, owning the layers in between from raw to curated, along with their schema, quality, and lineage.
- Build the data layer behind retrieval: source connectors, document parsing, chunking, embedding generation, and vector indexing, including re-embedding when content changes.
- Model curated, query-ready datasets and metrics so AI and analytics consumers work from one definition instead of each rebuilding the logic.
- Add quality checks, validation, and monitoring so problems surface before they reach a model or a user.
- Apply access control where it belongs: row and column level rules, PII handling, and entitlement-aware datasets, enforced as close to query time as the stack allows.
- Work with the platform and DevOps engineers to expose data and retrieval as documented, dependable services.
- Keep storage, compute, and query costs in check, with particular attention to the cost of embedding and vector workloads.
- Review code, write the documentation, and help shape how the team builds its data layer.
Requirements
- Eight or more years in data engineering overall. That includes hands-on work building data for AI or ML systems such as retrieval, embeddings, or feature data, which can be a more recent part of your background.
- Strong SQL and strong Python, including PySpark or similar distributed processing.
- Production experience across the AWS data stack: S3 for the lake, Glue for ETL and the Data Catalog, Athena for serverless query, and Redshift as the warehouse.
- Hands-on experience with a layered data architecture, whether you call it medallion (bronze, silver, gold), a data lake feeding a warehouse, or a lakehouse, including building the transformation stages that move data from raw to curated.
- Experience with an ELT or integration tool such as Airbyte, Fivetran, or Meltano, including building or maintaining connectors.
- Experience with event-driven pipelines using SQS and SNS, and with at least one streaming or change-data-capture technology such as Kinesis, Amazon MSK, or Debezium.
- Hands-on experience with a semantic or metrics layer over the warehouse, such as Cube or the dbt Semantic Layer.
- Hands-on experience with at least one vector store and embedding workflow: pgvector, Amazon OpenSearch, Pinecone, Weaviate, or Milvus.
- Comfort with columnar and open table formats: Parquet together with Apache Iceberg, Delta Lake, or Hudi.
- Working knowledge of an orchestrator such as Amazon MWAA, Step Functions, Dagster, or Prefect, and enough infrastructure as code to work closely with DevOps.