Team Meeting Bingo held 12 players and fell over at 500. The post-mortem: one DynamoDB partition key, about 250 joins a second, and how I fixed it properly.
Posts (15)
Everything I have written, newest first.
- StatusEvergreenLeveldeep-divePublished15 Oct 2026Reading time11 min readTagsawslambdadynamodbserverlessside-projects
- AWS gave Lambda a way to run your own web server and hold a connection open. I built a joke bingo app to find out what that's actually for.StatusEvergreenLevelintroPublished15 Oct 2026Reading time8 min readTagsawslambdaserverlessside-projects
- I'm not a developer and I've never made a game. I gave a coding agent a 392 character goal: port the Doom engine to TypeScript, unattended. It took 3.5 hours. That was the easy bit.StatusDatedLevelpracticalPublished10 Oct 2026Reading time15 min readTagsaiclaude-codeagentsside-projects
- I got tired of juggling YouTube, Moises and Songsterr to practise improvising, so I built my own on AWS. Here's the serverless architecture and what it actually costs.StatusDatedLevelpracticalPublished29 Sep 2026Reading time20 min readTagsawsserverlessarchitectureside-projects
- My agent memory system was eating 50-80k tokens before I'd typed a word. Here's how Obsidian CLI, Haiku subagents, and a file deduplication hook cut that by 80%. Part 4 of the AI Memory series.StatusEvergreenLevelpracticalPublished04 Apr 2026Reading time11 min readTagsaiobsidianclaude-codehookstoken-optimization
- Teaching AI agents to review their own session transcripts, catch lost decisions, and consolidate memory overnight. Part 3 of the AI memory series.StatusEvergreenLevelpracticalPublished01 Apr 2026Reading time11 min readTagsaiobsidianmcpclaude-codehooksdream
- CLAUDE.md instructions are advisory. Hooks are contractual. How I made AI agent memory actually work by adding deterministic enforcement to an Obsidian vault.StatusEvergreenLevelpracticalPublished18 Mar 2026Reading time11 min readTagsaiobsidianmcpclaude-codehooks
- Why I'm using plain markdown files for AI agent memory instead of vector databases and semantic search. Sometimes the boring solution is the right one.StatusEvergreenLevelpracticalPublished27 Feb 2026Reading time6 min readTagsaiobsidianmcpproductivity
- How a 500-line bash script evolved into an automated WAF assessment engine - and why deterministic tooling beats giving AI the wheel.StatusEvergreenLevelpracticalPublished17 Feb 2026Reading time12 min readTagsawsaiautomationwell-architected
- How skills transform AI assistants from context-amnesiacs into genuinely useful collaborators - and why the token savings matter more than you'd think.StatusEvergreenLevelpracticalPublished01 Feb 2026Reading time10 min readTagsaiclaudeproductivityskills