Indexing a year of video locally on a 2021 MacBook with Gemma4-31B (50GB swap)
Summary
A practical, first-person case study of indexing a year of video locally on a 2021 MacBook using Gemma4-31B, Claude Code, and multiple vision backends. The post details the end-to-end pipeline (from metadata extraction to YAML sidecars), a local-first architecture with no cloud reliance, and the lessons learned from hardware constraints, schema design, and iterative debugging. It argues that an effective index is the core prerequisite for any AI-assisted video editing workflow.