Building things, solving problems, and figuring out what AI can do

I'm a product builder who likes turning ideas into things that actually work. I'm exploring what happens when AI handles more of the implementation and I spend more time deciding what to build, solving problems, and testing ideas.

This is where I document what I build, what I learn, and what doesn't quite go according to plan.

What I’m building

The projects are probably going to be all over the place. That’s part of the fun. One week it might be a Raspberry Pi and some LEDs, the next it’s an AI-powered fantasy football experiment, figuring out how to make my home setup work from a vacation rental, or finding out whether a 12GB graphics card and a Raspberry Pi can run a language model well enough to be worth the trouble.

The common thread? I had a problem, an idea, or an excuse to build something—and AI made it a lot easier to see where it could go. Something usually broke along the way. That part’s in here too.

Spec by Screenshot: Building a Dividend Dashboard on Robinhood's MCP Server

David Woods published on 2026-10-01

Robinhood’s app is built for watching prices move. I wanted a page that answers a different question: what does this account pay me, and when? Most of the spec for that page turned out to be screenshots of Robinhood’s own app.

The Scan That Wouldn't Die: ClamAV on a NAS That Only Stays Awake 16 Hours a Day

David Woods published on 2026-09-24

I had never used Docker. This is how I used AI to learn it well enough to run a ClamAV virus scan on my Synology NAS, which goes to sleep at 11PM and gives every scan a hard deadline. Along the way: a silent 100MB scan cap, a scan that hung for an hour, and a few places where the AI needed checking.

The Obvious Answer Was Wrong (Local LLMs on a PC and a Pi)

David Woods published on 2026-09-21

I benchmarked open-weight LLMs on a Windows PC and a Raspberry Pi 5, and nearly every obvious answer turned out wrong: a GPU silently running on CPU, a ‘bigger’ model slower than no GPU at all, a judge model that graded a correct answer wrong. Local earned its place for narrow, lightweight work. It isn’t yet a replacement for the closed models I use every day, and I didn’t put them through the same tests.

/images/plex-marketing-stream-anywhere.png

Wagging the Dog: Convincing Plex My Vacation Home Is My Actual Home

David Woods published on 2026-09-18

My movie library lives on a NAS at home. Plex decided watching it from vacation should cost extra. Here’s how I taught my NAS to lie about its address — with a lot of help from Claude and one deeply misleading grep match.

/images/fantasy-football-meme-toilet-store.png

Building a Fantasy Football Co-Manager with Claude

David Woods published on 2026-09-14

Could AI make managing my family’s fantasy football league easier? I connected our Sleeper league to Claude, built an MCP server for FantasyPros, and ended up with an AI co-manager.

Random Colors Are Not a Bug

David Woods published on 2026-09-10

I powered up an LED strip with no data connected and it blinked random colors at me. That turned out to be normal, and it’s how I ended up building a Pi 5 LED controller with Claude.