London, United Kingdom
Senior Software Engineer at QuantumBlack, AI by McKinsey. I maintain Kedro, an open-source Python framework on the Linux Foundation’s LF AI & Data.
Five years on one codebase, in the open. I write the framework internals, the Spark and Databricks integrations, the React tool that draws your pipeline, and the proposals that decide where it all goes next, then I cut the release.
Selected work
Each of these is a real, linkable piece of engineering: the proposal, the pull requests, the trade-offs I got wrong the first time, and what shipped.
KEP-10 · Kedro Enhancement Proposal
Authored and shepherded a formal enhancement proposal through community review, a withdrawal, a redesign, and a Technical Steering Committee vote. Then built it.
Accepted by TSC vote
kedro-plugins #1185
A ground-up replacement for the most load-bearing dataset in the ecosystem, adding Spark Connect support and untangling a decade of Databricks path handling.
+1,956 lines, zero breaking changes
Shipped in Kedro 1.3.0
A validation subsystem that reads the type hints already on your node functions and turns them into runtime guarantees, delivered as six reviewable pull requests.
In every Kedro install since 1.3.0
Langfuse · Opik · AutoGen
The tracing and evaluation datasets that let a Kedro pipeline emit LLM telemetry the same way it writes a Parquet file, plus an MCP server that answers questions about Kedro itself.
~3,400 lines across two providers
Open source
Since October 2021 I have worked almost entirely in public, inside the kedro-org GitHub organisation. This is the breakdown.
The framework itself
Runner internals, the parameter validation subsystem, the `kedro new` tools flow, micropackaging, Rich logging, cloud `--conf-source`, and release engineering.
86 merged PRs ·643 commits
Datasets & integrations
SparkDatasetV2, Databricks and Spark Connect utilities, the vector-store abstraction, and the Langfuse/Opik LLM observability family.
44 merged PRs ·458 commits
The React visualisation tool
Flowchart model decomposition, DataCatalog 2.0 lazy loading, modular pipeline expansion, and the Redux preferences layer.
35 merged PRs ·487 commits
Project templates
The spaceflights, pyspark-iris and databricks-iris starters, including the canonical “initialise Spark via hooks” pattern.
17 merged PRs ·57 commits
The editor extension
Refactored the Python language-server validation architecture into a pluggable validator package with live catalog diagnostics.
9 merged PRs ·74 commits
Teaching material
Observability provider as an environment switch, and evaluation-pipeline training content.
3 merged PRs ·2 commits
| Repository | What it is | PRs | Commits |
|---|---|---|---|
| kedro-org/kedro | The framework itselfRunner internals, the parameter validation subsystem, the `kedro new` tools flow, micropackaging, Rich logging, cloud `--conf-source`, and release engineering. | 86 | 643 |
| kedro-org/kedro-plugins | Datasets & integrationsSparkDatasetV2, Databricks and Spark Connect utilities, the vector-store abstraction, and the Langfuse/Opik LLM observability family. | 44 | 458 |
| kedro-org/kedro-viz | The React visualisation toolFlowchart model decomposition, DataCatalog 2.0 lazy loading, modular pipeline expansion, and the Redux preferences layer. | 35 | 487 |
| kedro-org/kedro-starters | Project templatesThe spaceflights, pyspark-iris and databricks-iris starters, including the canonical “initialise Spark via hooks” pattern. | 17 | 57 |
| kedro-org/vscode-kedro | The editor extensionRefactored the Python language-server validation architecture into a pluggable validator package with live catalog diagnostics. | 9 | 74 |
| kedro-org/kedro-academy | Teaching materialObservability provider as an environment switch, and evaluation-pipeline training content. | 3 | 2 |
Named maintainer on the Kedro TSC, one of nine people with a vote on the direction of a Linux Foundation graduate project.
See the maintainer list →I also moved Kedro’s release process off CircleCI and onto GitHub Actions.
Talks & community
A public technical livestream for the Kedro community. I took over hosting it, which means booking speakers, shaping each session and running the broadcast. I present on it too.
Next up
Kedro Coffee Chat · August 2026
Presenting the KEP-10 design and a live before/after demo: a supplier feed with a duplicate ID, a malformed rating and an invalid flag, caught at the I/O boundary instead of deep inside a node.
GOSIM Paris 2026 · Station F
Invited talk at the Own Your Data Science & AI Workshop, hosted by scikit-learn, Probabl and Fraunhofer IAIS.
Kedro Coffee Chat · with Laura Couto
On GraphRAG, agents, and the unglamorous pipeline work that makes agentic systems reproducible. The most-watched Coffee Chat of the series.
Kedro community livestream
Took over hosting the community’s public technical stream: booking speakers, shaping each session, and running the broadcast. Recent episodes have covered the Data Catalog, spec-driven development, Kedro in VS Code, and evaluation pipelines with Langfuse.
Contact
The work I like best sits where research meets production: frameworks, developer tooling, and the unglamorous plumbing that makes data and AI systems reproducible. If that is what you are building, I would like to hear about it.