What is Decode the Stack?
Decode the Stack is a daily English-language technology briefing for software professionals. It is the sister publication to Decodifica.Tech, with a separate English edition, site identity, RSS feed, and podcast.
The English source catalogue contains 242 curated feeds spanning software engineering, cloud, DevOps and SRE, security, AI/ML, and technology news. Each edition links to the original reporting and adds context about the engineering decisions and trade-offs behind the headlines.
Who is it for?
The briefing is for software engineers, cloud engineers, SREs, security practitioners, technical leads, and other professionals who need a reliable picture of what is changing across the technology industry.
Why?
Technology does not have a shortage of coverage; it has a filtering problem. The same development can appear across many publications while useful signals are easy to miss.
Decode the Stack groups related reporting, selects stories for their relevance and coverage, and connects the details in one source-linked edition. The aim is to spend less time sorting through duplicate headlines and more time understanding what happened and why it matters.
How it works
The English publication uses the same 242-feed catalogue and core editorial
pipeline as its Portuguese counterpart. The maintained English configuration
is pipeline/sources.yaml.
- Collection: The scheduled pipeline reads public RSS and Atom feeds and filters items using category-specific recency windows.
- Clustering: URL and title deduplication, TF-IDF, and cosine similarity group overlapping coverage into a single story while preserving the original source links.
- Ranking: Source authority, coverage, relevance, category balance, and recency determine which stories receive deeper coverage and which appear as Quick Reads. Quiet days are not padded with weak stories.
- Contextual synthesis: Separate AI passes write the main analysis and Quick Reads in English. This is an original synthesis based on the supplied reporting, not a translation of the Portuguese edition.
- Podcast: A three-host English dialogue is generated from the source-linked synthesis and converted to audio.
- Publication: Hugo builds the daily article, RSS feeds, topic archives, search index, and statistics. A weekly Top 10 is synthesized from the accumulated story buffer on Fridays.
Publishing runs on the authenticated workflow in the Decodifica.Tech repository. AI credentials remain on that runner and are not stored in this public repository.
Technology stack
| Component | Technology |
|---|---|
| Feed collection | Python, feedparser, requests, and python-dateutil |
| Clustering | scikit-learn TF-IDF and cosine similarity |
| Contextual synthesis | OpenAI, Groq, and Azure OpenAI fallback providers |
| Podcast audio | OpenAI text-to-speech with Edge TTS fallback |
| Static site | Hugo Extended, HTML, CSS, and JavaScript |
| Hosting | Cloudflare Pages on the free pages.dev address |
| CI and publishing | GitHub Actions; the English repository’s CI only tests and builds |
| Comments | Giscus is prepared but disabled until the English repository authorizes the app |
No analytics or advertising trackers are added by the English publication.
Sources
The catalogue contains 242 configured feeds. Examples include:
- Software engineering: Netflix Tech Blog, Airbnb Engineering, Spotify Engineering, Stripe Engineering, GitHub Engineering, and Shopify Engineering.
- Cloud: AWS Blog, Azure Blog, Vercel Blog, HashiCorp Blog, Fly.io, and Supabase Blog.
- DevOps and SRE: Grafana Blog, Azure DevOps Blog, Kubernetes Blog, Docker Blog, CNCF Blog, GitLab Blog, and Brendan Gregg.
- Security: Krebs on Security, The Hacker News, Schneier on Security, Troy Hunt, Google Project Zero, Cloudflare Security, and Microsoft Security (MSRC).
- AI/ML: OpenAI Blog, Google AI Blog, Anthropic Blog, Hugging Face Blog, DeepMind Blog, Microsoft AI Blog, LangChain Blog, and Simon Willison.
- Technology news: Hacker News, Lobsters, Ars Technica, The Verge, TechCrunch, InfoQ, The New Stack, and TLDR.
The full, current list and its feed categories are available in
pipeline/sources.yaml.
Daily edition format
Each daily edition includes:
- Main stories: Contextual analysis of the selected reporting, grouped by technology category.
- Quick Reads: Concise notes on other relevant reports.
- Source links: Direct links to the original publishers and the reporting behind each story.
- Tags and related editions: Topics and entities connect coverage across the archive.
- Podcast episode: A conversational, three-host English audio edition with a target length of about 15 minutes.
The number of main stories and Quick Reads varies with the quality and volume of available reporting.
Site features
- Search: Search the published editions and their source-linked content.
- Archive: Browse daily and Friday weekly editions chronologically.
- Statistics: View published-edition totals, source coverage, and topic trends derived from the English archive.
- Weekly briefing: A Friday Top 10 re-ranks the stories accumulated during the week.
- Categories and tags: Browse by software engineering, cloud, DevOps & SRE, security, AI/ML, and technologies or companies.
- Related editions: Discover other posts connected by topics and categories.
- RSS: Separate feeds are available for written editions and podcast audio.
- Progressive web app: Install the site and access pages cached by the service worker while offline.
- SEO: Canonical URLs, JSON-LD, Open Graph metadata, sitemap, and robots directives are included.
Podcast hosts
| Host | Style |
|---|---|
| Anna | The curious, energetic lead host who guides the show and introduces each story. |
| Maria | The technical analyst who adds context, trade-offs, and thoughtful disagreement. |
| Alex | The relaxed co-host who asks listener-focused questions and brings occasional light humor. |
The script targets approximately 15 minutes. OpenAI text-to-speech uses the configured Marin, Sage, and Ash voices; Edge TTS is the fallback. The selected provider and model are reported in the publishing run.
AI and editorial safeguards
AI is used to synthesize context and write the podcast dialogue, not as a source of facts. Original publisher links remain with each story. The script is grounded in the source-linked edition, and the pipeline stops if all configured synthesis providers fail rather than publishing a headline-only fallback as contextual analysis.
Statistics and source counts describe the English catalogue and archive; they are not estimates of total technology-industry coverage.
Comments and feedback
Comments remain disabled while the English repository’s separate Giscus app installation is pending. No broken widget or Portuguese discussion identity is embedded. A podcast-directory listing link can be added once its official URL is available.
For corrections or suggestions, open an issue in the Decode the Stack repository and include the edition and original source URLs.
Read our privacy notice and terms of use.