100 days. 100 projects. One LLM prompt at a time.
Each post here is a project built with AI assistance — browser extensions, bots, tools, dashboards, analyzers. Some took an afternoon. Some fought back. All of them taught me something about what these tools are actually good at, and where they fall short.
The rule: build something real, ship it, write about it honestly.
A single-file browser tool that queries MalwareBazaar, VirusTotal, OTX and more — then uses Gemini to generate a production-quality YARA rule with a condition-by-condition explanation.
A single-file Cloudflare Workers app that stress-tests any claim, URL, or article from every angle — steelman, strawman, synthesis.
I built a timed communication drill for security leaders — 26 scenarios across three modes, with LLM coaching on jargon, clarity, and business framing.
I built a bot that posts P.G. Wodehouse quotes to Bluesky twice a day — similes on Mondays, dialogue threads on Wednesdays, character spotlights on Thursdays — rotating through seven formats across the week, with zero infrastructure.
I built a Flask web app that extracts, categorises, and archives every link ever shared in a WhatsApp group — 4,094 unique URLs across two years of conversation, now sorted, categorised, searchable, and downloadable as Excel.
A personal reader app for Adam Smith's Wealth of Nations — click any paragraph, ask Gemini questions about it in the full chapter context. Built in one session with Claude Code.
I scraped all 128 episodes of the EiE podcast and built reference spreadsheets for every book, song, and article recommended. Here's how.
A local Flask tool for APK security analysis — parses permissions, scans for secrets, fingerprints SDKs, and checks crypto usage. Nothing leaves your machine.
KaggaBot uses GitHub Actions as a free scheduler to post two Mankutimmana Kagga verses to Mastodon every day — no server, no cost, no babysitting.
Tried to build a home Icecast radio server to stream MP3s to my phone. What actually happened: four Docker containers, a Mega.nz dead-end, and a surprisingly working result.