DS DOOM_SCROLL Daily AI and ML signal desk
Archive RSS

GitHub Pages wiki

The zero-cost intelligence desk for research, agents, and ML systems.

DOOM_SCROLL pulls fresh AI and machine learning signals from public sources, ranks them with transparent heuristics, and publishes a clean daily briefing that looks professional without requiring paid infrastructure.

Open latest digest Browse archive Follow RSS
Coverage 6 public sources
Scoring model 5 weighted signals
Hosting $0 on GitHub
Latest issue

Daily AI Digest

The homepage stays stable while the issue pages, archive, and RSS feed update automatically after each run.

Waiting for site metadata Archive updates after each publish
Read issue See full archive
Free by design
$0 monthly platform spend
  • GitHub Actions runs the daily and weekly cron jobs.
  • GitHub Pages serves the archive and RSS feed.
  • Plain JSON outputs stay versioned in the repository.
  • Email delivery is optional and only needs Gmail app secrets.

What this page is

An editorial front door for the whole pipeline.

This root page is the permanent project guide: what DOOM_SCROLL monitors, how ranking works, what outputs are published, and how the whole stack stays tidy, reliable, and free to run.

Daily briefing

A ranked issue with deep dives, quick signals, and clean static HTML that is easy to read on mobile or desktop.

Machine-readable outputs

The same run emits Markdown, JSON, RSS, and run statistics so dashboards and automation can reuse the work.

Transparent logic

The ranking is heuristic-first: recency, engagement, cross-source confirmation, keyword relevance, and novelty.

Signal coverage

Six inputs, one digest.

Each source plays a different role: frontier papers, practitioner attention, code momentum, and long-form essays. The value comes from combining them, not from any single feed.

arXiv

Fresh research

Category queries across AI, ML, robotics, multi-agent systems, vision, language, and statistics.

Hacker News

Operator attention

Best and latest stories provide live signals about what builders are debating, sharing, and testing.

RSS

Curated essays

Selected feeds add commentary and applied engineering perspectives that raw paper feeds usually miss.

Hugging Face

Paper momentum

Surfaced papers get enriched and matched against arXiv so paper attention is not trapped in one ecosystem.

GitHub Trending

Code velocity

Trending repositories add implementation-level signals, especially for agents, diffusion, RAG, and infrastructure.

Papers With Code

Research to code

When the scraper lands usable results, this source helps connect papers to executable code and benchmarks.

Ranking model

Weighted, legible, and easy to tune.

The ranking model is not magic. It is a clear blend of timeliness, traction, confirmation, domain relevance, and novelty. That keeps the output inspectable and cheap to maintain.

Recency Inverse age decay keeps the front page current.
25%
Engagement HN activity and GitHub stars capture real-world attention.
30%
Cross-source confidence Items appearing in more than one source get promoted.
25%
Keyword fit TF-IDF style weighting favors the topics you actually care about.
15%
Novelty Rolling history avoids surfacing the same signal day after day.
5%

Publishing loop

From ingestion to a static page in one pass.

The system pulls, filters, enriches, scores, explains, renders, publishes, and stores history in one automated run. The archive grows while the root page stays polished and stable.

01

Ingest

Pull fresh items from arXiv, RSS, Hacker News, Hugging Face, GitHub Trending, and Papers With Code.

02

Filter

Apply source-specific time windows, quality thresholds, and novelty rules before ranking starts.

03

Enrich

Match HN discussions, merge cross-source duplicates, and add metadata that raises confidence.

04

Render

Publish HTML, Markdown, JSON, RSS, and a GitHub Pages archive that remains fully static.

Operations

Professional output without paid infrastructure.

The project is intentionally boring in the best way: one Python runtime, one static site, scheduled workflows, and optional email delivery. That keeps the maintenance burden low and the appearance high-end.

Static site

The homepage is the permanent guide. New digests publish as dated HTML pages, while the archive and site metadata update automatically.

RSS and HTML

Readers can subscribe by feed or browse individual issues directly. No database or app server is required.

JSON and history

The pipeline stores structured outputs and rolling history, which keeps scoring and dashboards grounded in prior runs.

Run locally

Use the existing uv-based workflow to generate the digest and refresh the static site.

uv sync --frozen
uv run python main.py

Automate for free

GitHub Actions handles the daily schedule, commits outputs, and keeps Pages updated from the docs directory.

schedule:
  - cron: "0 7 * * *"

optional email secrets:
  GMAIL_ADDRESS
  GMAIL_APP_PASSWORD
  DIGEST_TO_EMAIL