14 years in data engineering, 11 at Amazon across big data, fraud analytics, and AWS Finance. I design lakehouse platforms, global batch systems, and the GenAI layers that sit on top of them.
| years_experience | platform_users | glue_clusters | countries_served | teams_on_cdk_framework | interviews_conducted |
|---|---|---|---|---|---|
| 14 | 1,600+ | 25+ | 20+ | 12 | 33+ |
Designed and built a medallion-architecture lakehouse on Apache Iceberg serving 1,600+ users across 8 organizations, with row-level security so each team sees exactly the data it should — and nothing else.
Layered an embedded AI agent (Jarvis) on top of the semantic layer, letting finance users query 2,500+ datasets in plain language instead of writing SQL.
Built the multi-cluster batch system behind month-end close: 25+ Glue clusters orchestrated across 20+ countries, where a missed SLA means a delayed financial close for the business.
The hard problems were operational, not computational — idempotent reruns, cross-region dependency ordering, and making failures diagnosable at 2 a.m. without paging the whole team.
Built exactly-once-style streaming ingestion with Kinesis and Lambda, designed around idempotency so replays and retries never double-count financial events.
Packaged the team's infrastructure patterns into an AWS CDK framework that 12 other teams adopted — turning one team's hard-won conventions into an organization-wide standard.
A tool-using LLM agent over S&P 500 fundamentals: FastAPI backend, DuckDB store covering 503 tickers, an agent loop with 5 tools, and an eval harness to measure answer quality instead of guessing at it.
Own data platform architecture for AWS Finance: the AMPS Iceberg lakehouse (1,600+ users, 8 orgs), TCOL month-end close batch across 20+ countries, and an embedded AI agent over the semantic layer. Built the CDK framework adopted by 12 teams.
Built analytics and data pipelines supporting seller trust and support operations, turning operational event streams into decision-ready datasets for global teams.
Fraud and abuse analytics: built the data foundations that risk models and investigators depended on to detect bad actors across marketplace transactions.
Started at Amazon in the Big Data Technologies org, building large-scale batch processing on Hadoop-era infrastructure as the company scaled its internal data platform.
[Describe your first three years in data engineering before joining Amazon — company, what you built, and the stack.]
Apache Iceberg, Spark, AWS Glue, Redshift Spectrum, medallion architecture, semantic layers.
Kinesis, Lambda, idempotent ingestion patterns, exactly-once semantics for financial data.
Airflow / MWAA, AWS CDK, multi-region batch orchestration, infrastructure as code.
LLM agents over governed data, tool use, eval harnesses, natural-language access to semantic layers.
Fintech and FP&A: month-end close, journal entries, accruals, plus fraud & abuse analytics.
33+ technical interviews conducted at Amazon; mentoring, design reviews, and cross-team standards.