📰 Dev.to · MORINAGA

The two-speaker JSON spec format I use for AI-narrated YouTube explainers: segment beats, voice assignment, slide attachment, and what the CI build does with it

Building bluesky-qc.mjs: a four-gate pipeline stage that runs between a JSONL queue and the post script, catching vocabulary leaks, stale entries, and low-engag

Five open-source AI releases worth bookmarking from the week of May 26–30 2026, annotated for builders running lean AI stacks.

Weekly observations from running three programmatic SEO sites and an AI video pipeline: Actions budget, Bluesky post quality gates, CC0 image pitfalls, and more

Practical notes on using the Openverse API for license-clear slide photos: which license filter to request, why images fail silently, and what courtesy attribut

Pre-rendering Mermaid and matplotlib charts to PNG before slide composition is the only approach that works reliably in headless CI. Here is what that looks lik

Three open-source tools that cover the Datadog surface area — metrics, traces, and logs — what each actually does well, and which fits an indie dev's constraint

Building ossfind.com's data pipeline: two-phase seeding, GitHub API fetching, Claude Haiku summaries, and the DO NOTHING bug that silently erased polished conte

Comparing edge-tts, Kokoro-82M, MeloTTS, and Bark as free neural TTS options for automated video pipelines running inside GitHub Actions without a GPU.

Building a 1920x1080 PNG slide system with Pillow: dispatch tables, brand chrome, table layouts, and a thumbnail renderer — all from a JSON spec in CI.

How polish.py upgrades 380 model entries without an API call: MD5-based pool selection, license and framework extraction from HuggingFace tags, and why determin

The yt-analytics auto-tuner I run daily: median view classification, title-overlap archetype matching, first-word hook heuristics, and writing bias back to the

Five links from the week of May 19 that caught my attention as someone running three AI-curated directory sites on a $25/month stack.

A JSONL queue file, a Node.js script that builds richtext facets for hashtags and URLs, and off-minute GitHub Actions cron scheduling — no third-party scheduler

How I render 1920x1080 YouTube videos from a JSON dialogue spec using edge-tts neural TTS, Pillow slides, and ffmpeg concat — all inside GitHub Actions with no

RunPod, Vast.ai, and Hetzner Cloud as contextual affiliate links on model pages — how I implemented them, which ones are actually clickable, and what converts.

Using HuggingFace pipeline_tag to branch editorial copy at build time across 400+ model pages — what worked, what broke, and how I handle null tags.

How I slot per-entry human editorial takes into auto-generated comparison pages using Astro 5 content collections, and when the pattern is worth the overhead.

How I built methodology, about, and affiliate-disclosure pages for an AdSense re-application — and what surfacing code logic as prose actually means.

The isCurated gate I built for ossfind: generate all alternative pages, but gate which ones appear in sitemap, RSS, navigation, and robots meta.

How I set rate-limit sleep intervals for Steam, GitHub, and HuggingFace ETL pipelines — the specific numbers, the failure modes, and what each API actually tole

The model_used column, upgrade query pattern, and fallback chain I use to progressively enrich three Turso-backed Astro directories with Claude Haiku.

A practical comparison of Pagefind, Algolia DocSearch, and Lunr.js for an Astro SSG directory — build-time trade-offs, cost, and bundle size.

The O(n²) combinatorics problem with AI directory compare pages, how I budget-capped Claude Haiku calls, and dumped results to static JSON for Astro.