Data platforms
Lakehouse and warehouse architecture, migrations off legacy systems, governance built in from day one.
- Databricks
- Trino · Starburst
- Snowflake
- BigQuery
- Apache Iceberg
- Delta Lake
- Spark
- dbt
- Power BI
- Looker
Senior Technical Lead Engineer, Data Platform
I design and build lakehouse platforms that move data reliably at scale, and ship open source tools for the people who run them.
Platform architecture across the major engines, AI that works on real company data, and delivery pipelines that make both safe to change.
Lakehouse and warehouse architecture, migrations off legacy systems, governance built in from day one.
LLMs connected to governed data: AI functions in SQL, MCP servers for agents, and agentic frameworks that speed up migrations.
Everything as code and tested before it ships: PR-scoped environments, slim builds, autoscaling clusters and repeatable releases to Kubernetes.
A selection from platform work at Allianz Partners and BestSecret.
New company data platforms, on Databricks and on Starburst/Trino with Apache Iceberg.
Allianz Partners · BestSecretBestSecret: a Trino/Starburst and Apache Iceberg lakehouse on Azure, the migration off legacy platforms, ETL on dbt and Airflow, and Trino clusters autoscaled with KEDA on Kubernetes (write-up).
Allianz Partners: a Databricks lakehouse with dltHub ingestion and dbt transformation, plus agentic tooling to speed up the Oracle to Databricks migration.
A Databricks App that lets teams create controlled data exports instead of ad hoc extracts.
Allianz PartnersTeams need data outside the platform. Instead of one-off extracts, exports are created through an app that runs inside Databricks, so every export takes one controlled path.
Role and policy based access models for Starburst and Databricks.
Allianz Partners · BestSecretRole and policy based access models for Starburst and Databricks. Related writing: Trino with Open Policy Agent and OpenTelemetry.
Pull request builds that run and test only the models that changed.
dbt · DatabricksPull request runs compare against the production manifest and build only modified models and what depends on them (state:modified+), so CI stays fast as the project grows. Write-up: dbt Core and Slim CI on Databricks. Related open source: dbt-hodor.
A reusable ingestion framework on dlt for bringing new sources into the lakehouse.
Allianz PartnersIngestion on dlt (dltHub) that brings sources into the Databricks lakehouse at Allianz Partners, with dbt handling transformation downstream.
A Trino/Starburst plugin that exposes LLM calls as SQL functions, shown on Trino Community Broadcast 60.
Open source · trino-aiLLM calls exposed as SQL functions in Trino/Starburst, so models can be used from plain SQL. Code: trino-ai. Demo: Trino with Llama 3 on LinkedIn and Trino Community Broadcast 60.
Tools I built because the platform work needed them. Free, self-hostable, Apache-licensed.
Reuse unchanged dbt nodes and rebuild only when code or upstream data changed. A free, self-hosted alternative to dbt State, with Slim CI support.
A VS Code extension to design AI agents and skills on a visual canvas: templates, file handoffs, plugins and GitHub imports.
Claude Code mods for running many agent sessions across many repositories from one control tower.
Also: mcp-trino-python 25 GitHub stars, an MCP server for Trino · iceberg-metadata-insights 17 GitHub stars · trino-ai 16 GitHub stars, AI SQL functions for Trino.
Merged contributions to the projects my platforms run on: 9 pull requests across 6 projects.
| PROJECT | CHANGE | PULL REQUEST |
|---|---|---|
| Trino Gateway | JMX monitor type for cluster health checks | #472 #611 |
| Trino Gateway | Common labels for the Helm chart, UI API docs | #448 #471 |
| PyIceberg | Fix config loading on Windows | #7002 |
| Trino | Hive connector documentation | #22093 |
| Trino Charts | Consistent secret naming | #323 |
| Marquez | Common labels on ReplicaSets | #2877 |
| CloudBeaver | Trino driver upgrade | #2372 |
On calling large language models straight from SQL.
Field notes from running platforms in production, on LinkedIn and the BestSecret tech blog.
Fifteen years in data, from warehouse testing in regulated industries to running lakehouse platforms.
Allianz Partners
Databricks lakehouse architecture, ingestion and transformation frameworks with dltHub and dbt, agentic tooling to speed up the Oracle to Databricks migration.
BestSecret Group
Designed a Trino/Starburst and Apache Iceberg lakehouse on Azure, led the migration off legacy platforms, streamlined ETL with dbt and Airflow.
Commerzbank
Scalable big data solutions with Spark, Kafka and Google Cloud.
Accenture DACH
Test automation and data validation for large data warehouse programmes at UBS, Roche, Lufthansa, Aldi Süd and RWE.
Kraftfahrt-Bundesamt
Off the clock. From Bolu, Türkiye, where my favourite place is the house at Gölcük lake. I ride a Voge DS900X, on tours as far as the passes of the Italian Dolomites.