Spotify
AI Platform Infrastructure @ Spotify
Background
Four years ago I joined Spotify's Developer Experience team, working on the Backstage developer portal — the internal platform that thousands of Spotify engineers use every day. At the time, the team's focus was on plugin infrastructure and developer tooling. But as AI started reshaping how engineers worked, my role shifted along with it. I found myself building systems I'd never imagined: platforms for AI governance, marketplaces for AI skills, and the infrastructure that determines how an entire company discovers, trusts, and uses AI capabilities. Somewhere along the way I was promoted to Senior Engineer, but honestly the work itself changed more than the title did.
Solution
The biggest thing I built was the Unified Skill Marketplace — a product that went from an idea to company-wide infrastructure. The problem was simple: teams across Spotify were creating AI skills, rules, and MCP servers, but there was nowhere to find them, no way to evaluate their quality, and no standard for how they should work. I designed an entity system to bring AI context into the software catalog, built the backend APIs and frontend UX, integrated quality scoring so teams could trust what they found, and coordinated a cross-repo rebrand as the product evolved. It taught me that the hardest part of platform work isn't the code — it's designing something that makes sense for people you've never met.
Around the same time, I built the Rulebook platform from scratch: a way to model governance rules as first-class entities in the catalog. It was one of those projects where you're building infrastructure that other teams don't see but depend on — entity providers, catalog stitching, search, filtering, quality tracks. The kind of work where you know it's working when nobody notices it.
I also contributed significantly to AiKA, Spotify's internal AI assistant embedded in Backstage. I added multimodal image support, built MCP server selection with search, improved streaming reliability, and helped transition the product to its next major version. There's something satisfying about watching engineers across the company use something you helped ship, even if they never know your name.
One of the projects I'm proudest of started as a self-initiated prototype: AI-generated TechDocs. I wrote an RFC, built a proof of concept that analyzed code and generated documentation using LLMs, and eventually built a full pipeline that pushes generated docs to source repos via automated PRs. I also contributed a fix to upstream open-source Backstage, which felt like a small way of giving back to the ecosystem that our whole platform runs on.
What kept me going through all of it was the chance to build platforms that make other engineers more effective. That kind of multiplier effect — where your work enables work you'll never see — is what I love about infrastructure.