Data Engineer

Gunnar Morling Portfolio

Senior Principal Technologist, Confluent

Blog by Gunnar Morling, former lead of the Debezium change-data-capture project and founder of Hardwood, an open-source JVM library for reading (and, since 1.1.0.Beta1, writing) Apache Parquet files, with posts detailing its performance work and feature additions.

Change Data CaptureApache ParquetStreaming DataJava

Location

Hamburg, Germany

Companies

ConfluentDecodableRed Hat

Skills

Change Data CaptureApache ParquetStreaming DataJava

Notable

  • Led the Debezium change-data-capture project during ten years at Red Hat
  • Named a Java Champion
  • Ran the One Billion Row Challenge (1BRC): 8.1k GitHub stars, 179 contributors
  • Spec lead for Bean Validation 2.0 (JSR 380)

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

Deep-dive posts tied directly to a maintained open-source project

Posts on Parquet Bloom filters and geospatial support in Hardwood are written by the person actually building the library, so the technical detail reflects real implementation decisions rather than secondhand explanation.

2

A blog that tracks incremental releases, not just major milestones

Hardwood's previews each got their own post — Beta1, Beta2, CR1, then 1.0 in June 2026 and 1.1.0.Beta1 in August — rather than a single launch announcement, which keeps the writing cadence tied to real, ongoing progress.

3

Coverage spanning both the data-format layer and the systems layer

Posts move between low-level file format work (Parquet) and higher-level systems design (a durable execution engine, idempotency keys) — breadth that reflects working across the full data-pipeline stack, not just one layer of it.

What Data Engineers can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • Write about the exact version or feature you just shipped in your open-source project rather than only writing retrospectively about finished work.

  • Pair low-level implementation posts (file formats, protocols) with higher-level systems design posts to show you operate across the whole stack, not just one layer.

  • A blog that's a direct extension of an actively maintained project reads as more credible than one that only discusses other people's tools.

Projects

Gunnar Morling's blog doubles as a release log: in 2026 Hardwood got a post for nearly every preview on its way to 1.0 and beyond. The projects below read the same way, shipped, versioned and written up by the person who cut the release.

Open SourceadvancedOngoing since early 2026

Hardwood

Java 21Apache ParquetAmazon S3Apache AvroCLI/TUIMaven Central

A Parquet library for the JVM, written from scratch with no mandatory dependencies ("no Hadoop") and multi-threaded page decoding across all cores by default. Features added over the preview releases include an S3 backend, predicate pushdown, Avro bindings, VARIANT columns, GEOMETRY/GEOGRAPHY support, an interactive TUI for inspecting files, and, in 1.1.0.Beta1, write support, Bloom filters and dictionary-based row-group pruning.

It is file-format internals work (Dremel encoding, a fast path for fixed-length lists such as embeddings) shipped as a production library with release discipline.

  • Applies performance lessons from 1BRC

  • Every release written up on the blog

  • Apache License 2.0 on Maven Central

Open SourceadvancedJanuary 2024 challenge

The One Billion Row Challenge (1BRC)

JavaSIMDHigh-performance I/O

A coding challenge: aggregate one billion rows from a text file as fast as possible in Java. Run in January 2024, it became a crash course in modern Java APIs, SIMD and low-level performance for hundreds of developers, and spread to other ecosystems.

Designing a benchmark that a community wants to compete on — and adjudicating it fairly — is rarer than writing a fast solution.

  • Learnings later fed into Hardwood's design

  • Drew entries well beyond the Java community

ToolintermediateOngoing

kcctl

JavaQuarkusGraalVMKafka Connect

A command-line client for Kafka Connect: inspect the Connect cluster and individual connectors, and register, start and stop connectors. Built on Quarkus and shipped as native binaries for Linux, macOS and Windows via GraalVM.

It fills an operational gap in a widely deployed streaming component, the kind of tooling that platform teams depend on day to day.

  • Native binaries, no JVM needed at runtime

  • kubectl-style ergonomics for Connect

Ready to build your portfolio?

Follow the Data Engineer roadmap — skills, projects, and timeline to get hired.

Data Engineer Roadmap