How we process UK vehicle data using modern columnar analytics.
Chinese 125s fail 61% more often at year one, but the gap closes to 10% by year twelve. DuckDB queries on 874M DVLA records + MiniLM defect clustering reveal the mechanism: it's electrical, not mechanical.
Mar 2026 · 10 min read
How Polars and DuckDB power the CarHunch Score — a full rebuild of 128M records in under 2 minutes with no warehouse, no Spark, no distributed infra.
Feb 2026 · 8 min read
From MOT notes to insights: transform free-text defects into embeddings, cluster by meaning, and run semantic search with practical MLOps workflows.
Sep 2025 · 12 min read
A minimal reproducible pipeline using MiniLM and MLflow, with Docker-first workflow and practical experiment tracking patterns.
Sep 2025 · 9 min read
