MS Computer Science, George Mason University ('26) — I build data platforms and evaluate AI systems. My work sits where data engineering meets AI: getting messy data into trustworthy, governed shape, then using it to build and measure intelligent systems.
📍 Virginia, USA · Open to full-time roles and internships 🔗 LinkedIn
- Data platforms Databricks · Snowflake · AWS (Glue, Athena, Kinesis, Lambda, Step Functions) · Azure Data Factory · Delta Lake · PySpark
- Modeling & analysis SQL · Python · dimensional modeling (Kimball, SCD Type 2) · Power BI
- AI LLM/VLM evaluation · benchmark design · Claude API · prompt engineering
- Engineering practice Terraform · Git · pytest · CI-minded IaC
| Project | What it is |
|---|---|
| hb-ecommerce-lakehouse | Serverless lakehouse on AWS — Kinesis ingestion, Lambda schema validation, Glue PySpark medallion pipeline to a star schema on Athena, Step Functions orchestration, 100% Terraform, live dashboard. Measured 248× scan reduction. |
| databricks-retail-lakehouse | Medallion lakehouse on Delta Lake — Auto Loader incremental ingestion of 2.1M records, quarantine DQ gates, Kimball star schema, Unity Catalog lineage. Verified end-to-end run. |
| snowflake-governed-elt | Event-driven ELT — Snowpipe from S3, Streams & Tasks driving incremental MERGE, SCD Type 2 dimensions, three-tier RBAC and dynamic PII masking. |
| azure-medallion-online-retail-etl | ADF medallion ETL into Azure SQL — T-SQL MERGE for SCD2, data-quality gates that halt bad loads before reporting. |
| vlm-visualization-literacy | Deterministic evaluation of GPT-5.4 on the 6,000-image ChartX benchmark — 85.7% 2D accuracy, a 26-point drop on 3D charts, and six design guidelines that followed from the failure modes. |
| dns-anomaly-detection | PCAP analysis framework detecting DNS tunneling, cache poisoning, spoofing, and fast-flux botnets — entropy and TTL-based detectors with Claude API threat analysis and a Streamlit UI. |
Deepening the AI side of my portfolio — evaluation pipelines, retrieval systems, and applying LLMs to analytics problems — while keeping the data engineering foundation that makes those systems work on real data.