I'm a Data Engineer based in Ho Chi Minh City with 3+ years of building data platforms on AWS and Google Cloud. I like pipelines that are boring in production: idempotent, observable and easy to replay.
- 🔭 Working on ingestion that feeds a permission-aware RAG knowledge base on Amazon Bedrock
- 🌱 Exploring the modern data stack: Snowflake, dbt contracts, semantic layers
- 💬 Ask me about Kafka, Apache Iceberg, SCD Type 2 and Kimball modeling
- ⚡ Exactly-once streaming on Kafka and Apache Iceberg, safe under failure and replay
- 🚀 Cut a BigQuery pipeline from 5 to 3 minutes (40% faster)
- 🔎 Cut Elasticsearch search latency from 200 ms to 150 ms (25% lower)
- 🗂️ Built a 3-layer S3 Lakehouse with incremental, deduplicated ingestion
| Area | Tools |
|---|---|
| Warehouse & formats | Snowflake, BigQuery, Apache Iceberg, AWS Glue, Athena |
| Transform & orchestrate | dbt, MetricFlow, Airflow, Dagster, NiFi, dlt |
| Processing | PySpark, Spark Streaming, Kafka |
| Quality & governance | Great Expectations, dbt tests, OpenMetadata |
| Delivery | Docker, GitHub Actions, ArgoCD, AWS CDK |
End-to-end retail analytics: PostgreSQL → dlt → Snowflake → dbt → Preset, orchestrated by Dagster.
- SCD Type 2 dimensions in dbt, so every sale uses the product cost valid at that moment
- Metrics defined once in a MetricFlow semantic layer
- Great Expectations checks, retries, freshness warnings and Slack alerts
- Slim CI on per-PR Snowflake schemas and dbt contracts on every mart