Weekend Hack: Letting Ollama Sift 1,000 iOS Blogs for AI Gems

Over the last year I’ve had this itch to build again. AI made that possible: suddenly I can ship more small things, faster, and use them to push a long‑term plan forward-go from “mobile expert”, to “mobile + AI expert”, to someone who really understands AI as product, not just models.

The long game

That long game means learning how AI is actually being used in mobile development. Not the hype, This weekend’s step on that path was simple: if I want to learn how AI is showing up in the iOS world, I need to know who is actually writing about it.

Cue Dave Verwer’s treasure trove: the iOS Dev Directory ( ~1,000 blogs!), his iOS Dev Weekly newsletter, and the companion iOS Dev Jobs site. Too much for manual skimming, perfect for a weekend automation itch.

The tiny tool

I built a small CLI tool that pulls Dave’s blog list, fetches recent posts, and uses a local Ollama model to classify whether each post is actually about AI in a way that matters for iOS developers.

• Repo: gscalzo/iOS.Blogs.Analyzer
• Params: --months 3 (lookback window), --max-blogs 200 (how many blogs), --parallel 8 (threads), plus a --model flag to choose the Ollama model.
• Brain: local ollama model classifies each post as “AI‑related?”-fast enough, private, good enough.
• Output: a trimmed list I can drop straight into my Obsidian vault.

How it works under the hood

The architecture is deliberately boring: a small CLI that takes the parameters, pulls a list of sites from Dave’s directory export, and then runs a simple pipeline.

• Fetch: for each site (up to --max-blogs), download the RSS/Atom feed.
• Filter: keep only posts from the last --months months.
• Classify: send the title and a short excerpt to a local Ollama model with a tiny prompt: “Is this actually about AI, in a way that matters for an iOS developer?”
• Concurrency: process feeds in parallel (bounded by --parallel) so I’m not accidentally DDoS‑ing Dave’s list.
• Persist: write out a JSON/Markdown file with URLs, titles, and tags that Obsidian can pick up.

Most of the pieces came together through Copilot and Codex‑CLI suggestions: I described the pipeline, they sketched out the HTTP calls, concurrency, and file handling, and I iterated until the structure matched what I had in mind.

Why it felt magical

Saturday evening I was at the kitchen table, half keeping an eye on the oven timer, half watching logs scroll by. Copilot and Codex‑CLI handled most of the boring wiring; I mostly nudged, corrected, and decided what the tool should actually do. A year ago this would have been a “maybe next month” project. Now it comfortably fits between checking the roast and arguing about where the lights should go on the tree.

What’s next

• Swap models to compare local vs hosted.
• Add a “spice level” score: AI‑adjacent, AI‑heavy, AI‑marketing.
• Generate short summaries for each AI‑related post so I can scan themes quickly before deciding what deserves a deep read.

For now, I have a small, living map of AI conversations in the iOS community-and, more importantly, one more tiny brick laid on that long road from mobile to AI‑product work.