About NatSecPulse

NatSecPulse is the signal layer for defense technology, modernization & acquisition. It is not just a news reader: it ingests the open defense-tech firehose, clusters it into stories, extracts the companies and programs behind each one, scores the signal, and renders an explainable, buyer-grade feed at natsecpulse.com.

The intelligence layer

Under the feed sits a normalized entity graph: every story is linked to the companies, programs, and technology areas it touches. That graph powers dedicated Company and Program intelligence pages — each a rolling brief of who and what is moving, with the underlying sources attached.

The Buyer Dashboard turns that graph into a working surface: Company, Program and Technology radars, a Source Lens for who is breaking what, and an Acquisition Movement view that tracks contracts and programs over time. Every signal carries an explainable, per-signal score — relevance, confidence, and novelty are computed, not hand-waved — and everything maps to a standardized taxonomy of signal types, source categories, and tech areas.

Signal & trust

Weekly Report

A once-a-week synthesis of the most important defense-tech movements, with a web archive and its own RSS feed.

Source Leaderboard

Which outlets actually break and drive defense-tech stories, ranked by citation and lead time — not vibes.

Quality & Audit

A self-grading dashboard that audits the feed's own relevance, dedupe, and clustering quality so you can see where the pipeline is strong or weak.

Following

Track the companies, programs, and topics you care about and pull them to the top of your view.

Make it yours

Create an account with watchlists to follow specific companies, programs, and technology areas, and set up Saved Alerts — entity, acquisition, trend, and multi-source alerts — so the signal you care about finds you instead of the other way around.

For developers

Everything in the feed is available programmatically. Pull clusters from the REST / CSV / RSS export API, subscribe to the main RSS feed or the weekly RSS feed, and resolve labels against the canonical taxonomy.

How it works

The pipeline is built for a high signal-to-noise feed at low, predictable cost — no embeddings, and exactly one AI call per cluster.

  1. Ingestion — every 30 minutes, articles are fetched from trusted defense outlets and major news sources.
  2. Deduplication — near-duplicate articles are collapsed with content-based hashing.
  3. Heuristic clustering — related articles are grouped by Jaccard similarity over normalized titles. No embeddings, no vector store — fast and cheap.
  4. One AI call per cluster — a single Workers AI call (@cf/meta/llama-3.3-70b-instruct-fp8-fast) both classifies the cluster as defense-tech-or-not and, if in scope, writes the headline, summary, why-it-matters, and tags. Results are cached and never recomputed unless the cluster's contents change.
  5. Entity extraction + scoring — companies, programs, and tech areas are pulled out and linked into the graph, then each signal is scored for relevance, confidence, and novelty.
  6. Static, KV-cached render — pages are pre-rendered and served from Cloudflare KV for instant loads at the edge.

Tech stack

Cloudflare Workers (compute) · Cloudflare D1 (SQLite at the edge) · Cloudflare KV (pre-rendered HTML) · Cloudflare Cron Triggers (the 30-minute pipeline) · Cloudflare Workers AI (@cf/meta/llama-3.3-70b-instruct-fp8-fast, one call per cluster). Global edge deployment, sub-100ms reads, and costs that scale with stories — not users.