Production knowledge graphs,
end to end.

Cross-product relationships (add-on recommendations, compatibility, substitutions) were being curated manually, article by article. The process did not scale to the catalog, and the domain experts closest to the data had no way to express relationships as rules.

Built a reasoning application on RDFox Datalog and Metaphactory that lets home-furnishing knowledge engineers author cross-product relationships as declarative rules. Each rule generates thousands of pairings automatically. Rearchitected the Databricks pipeline so the domain team contributes directly to the KG, with validation errors surfaced back to the authors. Delivered the ontology, mappings, and the Azure platform underneath.

Manual article-by-article curation replaced with rule-based authoring controlled by the people who own the domain. Thousands of product pairings produced per rule, driving measurable lift in online conversion.

Millions of engineering catalog terms needed curation by engineers across the business, with full versioning and auditability. The existing tooling did not give domain teams a safe way to edit taxonomies, and there was no way to see how the catalog had looked at a given point in time.

Designed and built an internal Metaphactory-based application, acting as sole developer and maintainer. Shipped a custom enumeration editor and a time-machine interface that lets engineers browse any historical state of the catalog. Lineage and auditability are modeled on PROV-O. Also owned the supporting AWS infrastructure end to end: Collibra on EKS, deployments via AWS CDK and Terraform, and resolution of cross-team issues (Graviton RDS migration, Terraform and EKS version drift, Route 53 and ALB DNS misconfigurations).

Domain engineers curate a catalog at the scale of millions of terms, with a navigable history and full provenance. One integrated system where there used to be several.

RINF is the EU-wide regulatory system tracking rail infrastructure across every member state. The platform had grown into a tangled set of microservices that the agency could no longer evolve internally, and its semantic transformation pipeline was too slow to keep up with member-state submissions.

First deployed the legacy version to production, then led the full rearchitecture: replaced the microservice tangle with a modular FastAPI application on EKS Fargate. Rebuilt the semantic transformation pipeline from scratch in Python on an asynchronous Celery and SQS architecture. Owned authentication and authorization end to end via Azure Active Directory.

50x speedup on the semantic transformation pipeline. A maintainable codebase the agency can evolve internally, running EU regulatory infrastructure in production.