Updates April 6
PGrid
- started on RAM descriptions. created/iterated on prompt for generating the description. updated the verification part to be able to respond to Perplexity output stating what the issue is with the output and to regenerate the markdown file if required.
- even after this complicated review step there were still inaccuracies according to codex which did a review on final outputs.
- had further intermittent Tailscale connection issues with hosthatch2 server, migrated to a new temporary server on onidel then moved the hosthatch2 server to a new region and migration over again. issues seem to have stopped after this.
- working on a new social media post but there were multiple issues with the AI creating the slides. worked with another AI to iterate on the documentation around the social media lab to make content creation more consistent and behaviour the same across AI. created a new slideshow comparing US retailers for GPUs which I’m happy with.
- added skills to find newly released CPUs/GPUs and then directly run the specs search job for those if there are items in queue. added new CLI command to run single specs search jobs so AIs that find new items can run those jobs manually. used this to create 3 new CPU models for recently released Intel CPUs.
- updated eBay scraper to handle cases where ‘seller is away’ and return out of stock for these items.

- updated eBay scraper to start ignoring stores based on rating percentage as well after noticing that the store had a very bad percentage. before it was just based on number of ratings
- updated classification TUI so I can run the dual small LLM consensus from the TUI directly
- ran through an item audit session with a Codex agent, it incorrectly reviewed an item as a reclassify action where instead of reading the title it based its reasoning on a part number. updated instructions for this case, got the agent to create a skill so agent reviews can be done more autonomously after a database-wide audit is complete.
- work in progress on laptops, created the tables required for storing laptop information. started trying out Codex SDK to be able to run Codex after a specs search session is complete to allow it to make final review and judgements for each item by searching the database first and having access to search-based LLMs if needed.
- tried to use RapidAPI for AliExpress data but pricing information was incorrect, and search results were very inaccurate. also found from testing that AliExpress displays different prices quite often between logged in users and ‘new’ users. it can also change prices after refreshing a single listing page as a logged out user. for Australia it also doesn’t display taxes until checkout page. AliExpress would be quite hard to scrape