TEK2day Pulse

Read TEK2day Pulse for the latest technology news.

Live site: pulse.tek2dayholdings.com · JSON feed: pulse.tek2dayholdings.com/pulse.json

TEK2day Pulse is an automated technology news site published by TEK2day Holdings. It pulls from 21 curated RSS feeds, removes duplicates, keeps the stories that are about AI, software or fintech, and publishes a static site and a JSON feed hourly during weekday business hours.

What it publishes

• A single chronological front page: a lead story, a five-story Brief, and a grid of the latest stories with search
• pulse.json, a public JSON feed of the same stories
• A permalink page for every story, with a social-sharing image
• Daily and timestamped JSON archives

What gets in

The feeds are general technology publishers, so not everything they print belongs here. Each story is checked against vocabularies for AI, software and fintech; a story that matches none of them is dropped. On a typical day that removes car launches, game releases, gadget reviews, conference promotions and entertainment pieces. Everything dropped is listed as [DROP] in the build log so the filter can be audited.

Matching is on whole words (with plural and verb endings), weighted title 3×, summary 2×, URL 1×, and a story needs a score of at least 2. Everything from PYMNTS, Cointelegraph, OpenAI, Anthropic, the TEK2day newsletter and the configured YouTube channels is kept without scoring. Shopping and deals posts are removed separately, and Hacker News and X/Twitter links are blocked.

Each edition carries at most 10 stories per source (max_items_per_source in generator/config.yaml; the TEK2day newsletter is exempt), so no single feed can dominate the page.

Sources

Configured in generator/config.yaml:

• TechCrunch, The Verge, Ars Technica, Wired, VentureBeat
• Bloomberg Technology, WSJ Technology, The New York Times Technology, Financial Times Technology
• NVIDIA, Microsoft and Apple newsrooms; OpenAI News; Anthropic News
• PYMNTS and Cointelegraph
• The TEK2day newsletter
• YouTube channels: T2D Pulse, OpenAI, Anthropic, SS&C Blue Prism

Feeds that fail on a run are logged and recorded in build-status.json; the healthy ones publish as normal. (Anthropic's news feed and two of the YouTube feeds currently return errors and contribute nothing.)

The page

Header with logo, date and search. Below it, a hero with the lead story — the Brief's top pick when it has a real image, otherwise the first pick with a thumbnail, otherwise the newest story — beside the Brief, five stories ranked by impact. Then Latest: every story, newest first, with the build time in the section head. Typing in the search box filters the grid and hides the hero; clearing it brings the hero back.

Update schedule

GitHub Actions builds and deploys the site:

• Weekdays: hourly runs (23 8-17 * * 1-5, timezone America/New_York) from 8:23 AM through 5:23 PM Eastern, with automatic daylight saving adjustment. Minute 23 avoids GitHub's congested on-the-hour slots. GitHub schedules are best-effort: runs can be delayed or dropped, so these are target times rather than guaranteed publication times. Hourly attempts provide another opportunity to refresh after a missed trigger.
• Weekends: no scheduled builds; Friday's edition stays live. generator/check_schedule.py enforces this. Set PULSE_WEEKDAYS_ONLY to "0" in the workflow to publish seven days a week.
• On merge: every push to main builds and deploys immediately.
• Manual: Actions → Build TEK2day Pulse → Run workflow builds any day.

To check freshness, compare generated_at in the live build-status.json with the workflow run history. A green push or manual run verifies the build and deployment, but does not verify that scheduled triggers are firing. If no scheduled run appears for more than two hours during the publishing window, trigger a manual build and investigate the scheduler.

Builds are incremental: story pages, images and archives persist between runs in the Actions cache, so a run only fetches and encodes new stories (typically well under a minute). A run that yields no stories fails before deploying, so the previous edition stays up.

JSON feed

pulse.json is public and needs no authentication. It is an array of stories, newest first:

[
  {
    "title": "Article headline",
    "url": "https://publisher.example/story",
    "published_at": "2026-09-12T20:15:00-04:00",
    "source": "The New York Times - Technology",
    "summary_text": "Plain-text excerpt",
    "image_url": "https://publisher.example/image.jpg",
    "category": "ai",
    "_permalink": "/p/3cf421aa7d/",
    "_abs_permalink": "https://pulse.tek2dayholdings.com/p/3cf421aa7d/",
    "_summary_240": "Excerpt trimmed for cards",
    "_thumbnail": "https://pulse.tek2dayholdings.com/p/3cf421aa7d/thumbnail.jpg"
  }
]

category records which vocabulary the story matched (ai, software or fintech). It is kept for API consumers and is not shown on the site.

import httpx

stories = httpx.get("https://pulse.tek2dayholdings.com/pulse.json").json()
latest = stories[:10]

Kilby AI reads this feed through its search_t2d_pulse tool.

Project structure

generator/
  generate_pulse.py       Pipeline: fetch, dedupe, filter, render, archive
  check_schedule.py       Weekday gate for scheduled workflow runs
  preview_pulse.py        Render a saved pulse.json without network access
  config.yaml             Feeds, blocked domains, per-source cap
  requirements.txt        Python dependencies
  templates/
    section_template.html Front page
    item_template.html    Story permalink page
tests/
  test_pulse.py           Regression suite (unittest)
.github/workflows/
  tek2day_pulse.yml       Build + deploy (schedule, push to main, manual)
  checks.yml              Tests on push and pull requests
docs/                     GitHub Pages site
  pulse.css, pulse.js     Front-page styles and behaviour
  sw.js, offline.html     Service worker and offline fallback
  index.html, pulse.json  Generated each build (not tracked)
  build-status.json       Generated: feed health, counts, image reuse
  p/                      Generated story pages, og-image.jpg, thumbnail.jpg
  archive/                Generated daily and timestamped JSON snapshots

Tech stack

• Python — pipeline, deduplication, relevance filter, image processing
• feedparser — RSS/Atom parsing (fetched through a retrying HTTP session with timeouts)
• BeautifulSoup4 — HTML parsing
• Pillow — image resizing; social cards and thumbnails are progressive JPEG
• GitHub Actions — scheduling, incremental cache, deployment
• GitHub Pages — static hosting

Local development

Python 3.11+ with generator/requirements.txt installed in a virtual environment. A full build fetches the feeds and generates images; it runs on Windows, macOS and Linux:

python generator/generate_pulse.py
python -m http.server 8765 --bind 127.0.0.1 --directory docs

A second run is near-instant because images already under docs/p/ are reused; delete docs/p/ to force a full regeneration. docs/build-status.json shows which feeds succeeded and how many images were reused versus generated.

To work on the page without fetching anything, save the public feed to .tmp/live-pulse.json and render it:

python -m generator.preview_pulse --feed .tmp/live-pulse.json
python -m http.server 8765 --bind 127.0.0.1 --directory docs

Tests:

python -m unittest discover -s tests -v
node --check docs/pulse.js
node --check docs/sw.js

The suite covers the relevance filter (whole-word matching, off-topic drops, publication routing), feed errors and timeouts, safe rendering of feed content, the weekday gate, per-source capping, image reuse, sitemap dates, archive merging, preservation of the last edition when feeds fail, and a complete build with mocked network input. GitHub Actions runs it before every deploy and on every pull request.

License

Proprietary — © TEK2day Holdings. All rights reserved.