Updates August 8

PGrid

• Continue working on ‘live’ SSD classification. Have an initial brand classification that only focuses on extracting the brand from an SSD (or ignoring it if not SSD). Added pipeline to then classify SSD model then SSD variant.
• Updated brand classification session tracking to have a single session.
• After running brand classification, this is the resulting items that have brands from each source. There are 4000 SSDs to further classify. From that the brands are distributed like:
• Running Claude agent to manually look through classified SSD brands and find most popular models and see if we have those models in the database and if not queue a model spec search job.
• Did a refactor of how specs searches were done. Had 2 queues one for brand one for variants of SSDs but realised it would be better to have 1 configurable queue. Required a bit of refactoring but wanted to get it out of the way.
• New CLI for pgrid to allow checking of items quickly both for me and for Claude.
• Created a Claude slash command which uses GitHub CLI and fixes stuck frontend builds when the frontend gets stuck waiting for a self-hosted worker.
• Started refactoring spec searches to be routed from a single file. It can go from specs search queue to different destinations.
• Lots of iterating on SSD classification. This is how the model + variant side of classification looks like with the ability to queue spec searches. Still not able to run without issues.