The engineering team behind GFoundry
Engineering,
in practice.
We write about what we built, how we built it, and what we learned. Including what went wrong.
Explore the writing-
Latest article
Fast to produce is not the same as safe to follow
Read the article →: Fast to produce is not the same as safe to followWe gave our clients AI tools to write their own training. In regulated industries, that also handed them a new way to be wrong. How we built Gi Guardian to catch dangerous…
More writing
All articles →-
Zero is not the same as nothing
A client opened an export from our platform and found a column of zeros. The real numbers existed – the system just went to an empty table and wrote down what it…
-
Managing HR from Your LLM of Choice – GFoundry’s MCP in Practice
GFoundry shipped an MCP server. The interesting part is not the technology – MCP is straightforward enough. The interesting part is what becomes possible when your HR platform sits in the same…
-
The platform isn’t dying. It’s changing who it talks to.
Some people argue the era of the software platform is over, and that we’ll soon do everything inside one large language model. I think that’s half right. The work doesn’t disappear. It…
Different products.
The same curiosity.
The decisions behind the code are more interesting than the code itself.
Our storyFinding the next song.
A music social network, a recommendation engine, and the challenge of understanding what each listener wanted to hear.
Read the engineering story →Learning from behaviour.
Stream mining infrastructure built in an R&D partnership with Fraunhofer Portugal, and the work that followed.
Read the project retrospective →Understanding people at work.
A multi-tenant SaaS platform, years of decisions, and the lessons that only arrive after something has been running for a while.
Read the architecture retrospective →From the archive
Browse the archive →Ten years of multi-tenant SaaS: the decisions we made, and the ones we’d unmake
GFoundry. What held for ten years, what didn’t, and what a small team learns when decisions made in year two still cost you in year eight.
How we built a music recommendation engine in 2007 – before Spotify existed in Portugal
Palco Principal. 70,000 tracks, three engineers, no library worth using. The algorithm, the A/B test numbers, and how the 2008 financial crisis ended it.