← Polymath.OS // Captain's Log
Six Weeks in Review
2026-08-19
date: 2026-08-19
Six weeks of recap; everything from early July to mid-August — what got built, what got learned, and a lot of ideas. Still need to make it less cringe from Claude
The site (this thing)
Most of the through-line was Polymath.OS — the site you're reading. It went from a notes-sync script into a full portfolio OS: a push-to-publish pipeline (jot a note → it becomes a post → the site rebuilds), the Skill Tree with ~1500 skills including a recursive nesting layout Claude one-shot, the Captain's Log, a Project Matrix, first-party analytics, comments, and a home-page Growth Orbit that maps my skill XP as moons circling a planetoid. Late in the stretch I added live, editable theme + growth configs — anyone can retune the colors, fonts, and the orbit from inside the site, and I can publish my version for everyone.
The project spree
Somewhere in here the floodgates opened and I started shipping small AI-built projects faster than I could name them:
- Photo-editing SaaS — an automated editing pipeline, pushed to ~90%. Two SaaS products almost done at one point.
- YouTube lyric translator — pulls a video, translates lyrics with AI. Actually helping me chip away at Korean.
- A pile of language apps, including a live, gamified travel-translation one I think has legs.
- Promptrick — a clipboard monitor for AI prompts, with an Orchestrator agent to chain outputs between chat windows.
- An AI project-management harness / local browser agent to
Write-ups & research
The posts that were actually guides:
- Brave high-CPU fix — traced laggy YouTube + dropped frames to Brave silently switching to ANGLE D3D11-WARP (software rendering), and wrote up the diagnosis + fix.
- Filesystem memory for LLM agents — a practical breakdown of the arXiv paper on how agents organize markdown memory trees.
- A kratom-safety research site — compiling the actual science, with the message that mixing is the real danger, not the substance in isolation.
Machines & infrastructure
A recurring saga: burning through Claude limits, a bloating Windows install begging for a reinstall, standing up a Linux box (a three-hour LAN-mouse fight), and moving toward sandboxed, containerized coding (WSL/Docker) so one bad AI script can't fuck my whole machine.
Ideas & rabbit holes
The logs are half build-log, half notebook. Threads I kept pulling: a parallel-worlds reading of time travel after re-watching Tenet; higher dimensions folded into each other; "noble houses for compute" as a way to hand out AI access; the paperclip-maximizer and why a reasonable AI takeover is a choice, not an accident; a long critique of parasitic jobs and where work goes as AI eats the boring 90%; and a running fascination with dreams — cross-dream references, and spaces around my house that don't exist when I'm awake.
What's next
Publish and actually advertise the lyric app, finish the photo SaaS, keep the airship design alive, tighten the AI workflow so the projects stop half-finishing — and keep the daily log going, now that the Growth Orbit finally has something to show for it.