SignalScope

Live brand and market monitoring for GTM teams. SignalScope takes a brand, competitors, market category, and monitoring objective, then pulls fresh public signals, triages the noise with AI, and writes a cited market-intelligence report with concrete GTM recommendations.

It is designed for the gap between raw alerts and useful strategy: collect the signals, filter what matters, connect them to action, and keep every claim traceable to evidence.

Preview

What It Does

1. Collects live signals from Google News RSS, Hacker News, GDELT, and optional Tavily search.
2. Scores relevance with a fast AI triage pass so homonyms, spam, and off-topic results drop out.
3. Synthesizes a report with executive summary, sentiment, themes, competitor moves, opportunities, risks, and action items.
4. Cites every claim with evidence chips that jump back to the source signal.
5. Exports Markdown for sharing or follow-up analysis.

Business Applications

SignalScope is a reference pattern for turning noisy public information into decision-ready intelligence. The same architecture can support:

• Competitive intelligence: monitor launches, pricing changes, positioning shifts, partnerships, and hiring signals across a market.
• Brand and reputation monitoring: separate meaningful coverage from keyword noise and summarize changes in sentiment or narrative.
• Sales and account planning: watch named accounts, competitors, and industry events so reps can personalize outreach with current evidence.
• Founder or investor research: track emerging categories, company momentum, and market risks without manually checking multiple sources.
• Customer success risk detection: monitor public complaints, outages, competitor mentions, or regulatory events that may affect key accounts.

The core transferable idea is a source-grounded intelligence pipeline: collect broad signals, filter them cheaply, synthesize only the relevant evidence, and keep the final recommendation auditable.

Example Output

A generated report includes:

• Executive summary: what changed, why it matters, and what the GTM team should do next.
• Signal themes: grouped market, brand, competitor, and customer signals with citations.
• Competitor moves: notable launches, positioning shifts, partnerships, or public momentum.
• Opportunities and risks: practical recommendations grounded in the collected evidence.
• Evidence panel: source links and citation chips so reviewers can inspect the underlying signals.

Pipeline

Brief
  brand + competitors + category + objective
        |
        v
Collect
  Google News · Hacker News · GDELT · Tavily
        |
        v
Triage
  fast model scores relevance and removes noise
        |
        v
Synthesize
  analyst model writes a schema-checked cited report
        |
        v
Review
  report + evidence panel + Markdown export

Architecture

Layer	Implementation
App	Next.js App Router, React, TypeScript
Pipeline	lib/pipeline.ts orchestrates collect → triage → synthesize
Sources	Source adapters in lib/sources/
AI providers	Anthropic preferred, OpenAI fallback through Vercel AI SDK
Structured output	Zod schemas for requests, signals, stream events, and reports
Streaming	NDJSON progress events from app/api/monitor/route.ts
Evidence	Server strips invented citation IDs before the UI renders claims

Transferable Implementation Patterns

• Two-pass AI pipeline: a fast triage step reduces cost and noise before the higher-quality synthesis step.
• Adapter-based source ingestion: each signal source is isolated behind a small adapter, making it straightforward to add domain-specific feeds later.
• Schema-checked reports: Zod schemas define the shape of requests, stream events, signals, and reports so the UI can trust the API contract.
• Auditable recommendations: the report is only useful if a reviewer can inspect the supporting signal; citation filtering happens server-side before rendering.
• Progressive streaming UX: users see collection, triage, and synthesis progress instead of waiting on a silent long-running request.

Run Locally

npm install
cp .env.example .env.local
npm run dev

Open http://localhost:3000, use the example brief, and run monitoring. A typical run takes 30-60 seconds.

Smoke Test

With the dev server running:

npm run smoke
npm run smoke -- "Notion" "Coda,Airtable" "productivity software"

The smoke test streams pipeline events, validates that a report was produced, and checks that citations point to delivered signals.

Environment Variables

Variable	Required	Purpose
ANTHROPIC_API_KEY	One AI provider key required	Preferred provider for triage and synthesis
OPENAI_API_KEY	One AI provider key required	Fallback provider when Anthropic is not set
TAVILY_API_KEY	No	Adds web search as a fourth source
TRIAGE_MODEL	No	Override the triage model
SYNTHESIS_MODEL	No	Override the synthesis model

Deploy

npm run build
npx vercel

Set ANTHROPIC_API_KEY or OPENAI_API_KEY in Vercel. Add TAVILY_API_KEY if you want broader web search.

Engineering Notes

• Keyless-first collection. The default sources work without paid scraping or authenticated social APIs.
• Two-stage model use. A cost-efficient triage model filters noise before the synthesis model writes the final report.
• Citations are enforced in code. The UI only renders citation IDs that exist in the delivered signal set.
• Source failures are isolated. One source can fail without breaking the whole run.
• No fake data on empty runs. Thin evidence returns coverage notes and guidance instead of invented findings.

Known Limits

• No X, LinkedIn, Instagram, or Reddit ingestion without authenticated or paid APIs.
• News snippets are shallow; a production version would fetch and summarize full article bodies for the top signals.
• Reports are session-scoped and export-only; there is no saved history or week-over-week diffing.
• Recency windows vary by source and are tuned for useful signal volume.

License

MIT