How it works
How Radar becomes the trigger layer for your analytics stack
536 sources ingested hourly. Extracted against 1,000+ MA-specific keywords. Scored against your entity watchlist, geography, and regulatory calendar. Routed to your team ranked by urgency. Then you decide what to act on — with hours of lead time instead of days.
Hover each stage to see details
Ingest
Continuous multi-source collection
RSS feeds, CMS.gov filings, Federal Register, SEC EDGAR, state DOI bulletins, provider newsrooms, and Brave Search API — scanned hourly.
Extract
1,000+ keyword taxonomy + entity recognition
Two-tier keyword matching (primary + modifier) across MA-specific categories. Simultaneous entity resolution against a curated entity database with alias and hierarchy support.
Score
12-component relevance engine (0–100)
Source tier authority, keyword density, entity match strength, client market overlap, recency decay, CMS calendar alignment, seasonal adjustment, and penalty deductions.
Route
Watchlist & geography matching
Signals are matched to your entity watchlist, competitive positioning, and county-level service area. Only the intelligence that affects your markets reaches your team.
Brief
AI-generated executive summary
Domain-specific LLM summarization per signal category. Priority-grouped, impact-assessed, and delivered as a 3-minute morning brief — dashboard, email, or Slack.
Ingest
The pipeline ingests from 536 sources hourly using three parallel methods: RSS parsing for structured feeds, web scrapers for sites without RSS, and API integrations for programmatic sources like Brave Search and SEC EDGAR.
Every article is deduplicated at ingest using a content hash. The same press release syndicated across 12 outlets enters once. Articles are stored with full raw text, publication metadata, and source provenance for traceability.
Source authority tiers
Extract
Keyword taxonomy
Every article passes through a 1,000+ keyword taxonomy organized into MA-specific categories. Keywords are classified as “primary” or “modifier”. The word “network” alone is noise; “network termination” is signal. Porter stemming normalizes forms so “terminating,” “terminated,” and “termination” all match.
Entity recognition
Simultaneously, articles are scanned against 5067 known entities using fuzzy matching with alias support. UnitedHealthcare, UHC, United Health Group, and Optum all resolve to the same parent. Headline mentions score higher than body mentions. Entity-signal relationships are stored for routing.
Score
The scoring engine produces a relevance score from 0 to 100. Entity match is weighted heaviest (40 points) because in MA, who is involved matters more than what happened. A network provider event at your largest competitor in your top market scores higher than regulatory guidance outside your footprint.
Source tier (government filings boost score), keyword density, geographic overlap with your service area, regulatory calendar alignment (CMS deadlines amplify score), and market relevance all contribute. Signals scoring 81+ become P0 alerts — delivered immediately so your team can assess impact and respond.
Route
Scored signals are matched to your watchlist: self (your organization), competitors (direct rivals), and market (general participants). A signal about Humana's network change in Harris County, TX reaches only organizations operating in Harris County with Humana in their competitive watchlist. Noise is eliminated before it reaches your team.
P0 critical alerts (81–100) trigger immediate email and webhook delivery — your team gets the structured data they need to respond fast. P1 and P2 signals (41–80) batch into the morning brief, grouped by category with summaries. P3 signals (21–40) appear in the dashboard feed for context and long-term pattern recognition.
Brief
The final stage generates summaries and impact assessments. Each signal category has a dedicated prompt template: regulatory compliance emphasizes deadlines and affected entities; network provider focuses on geographic impact and estimated lives. The goal is a summary your team can act on without reading the full source article.
The morning brief assembles overnight signals into a prioritized, grouped view structured for fast executive scanning. One click drills into the source article. Another opens the full signal detail with entity context, geographic tags, and related signals.
See the pipeline in action
The pipeline has processed 112,997 signals so far. Request a walkthrough and see what it found today.