Signal Engine

Automated federal opportunity intelligence. Monitors SAM.gov solicitations and OIG/GAO watchdog reports daily, scores each item against your company profile, and surfaces ranked triage cards on a private dashboard.

Built for government contractors who want to stop missing opportunities because they posted on the wrong day.

How it works

1. Daily ingestion — cron jobs pull from SAM.gov (contract solicitations), OIG (audit reports), and GAO (watchdog reports)
2. Two-layer fit model — hard gates first (NAICS whitelist, contract size range, exclusion keywords), then LLM capability scoring against your config/profile.yaml
3. Triage cards — each passing item gets a BLUF, a solution hypothesis grounded in your proof points, a score (0-100), and a recommendation: pursue / review / track / skip
4. Private dashboard — passphrase-protected Next.js app; thumbs up/down feedback tunes the model over time

Stack

• Next.js 16 (App Router)
• Drizzle ORM + Neon Postgres
• Vercel AI SDK (Google Gemini by default)
• Vercel Cron + OIDC auth
• Tailwind CSS

Setup

1. Clone and install

git clone https://github.com/michaelrobertsutton/signal-engine.git
cd signal-engine
npm install

2. Configure your profile

cp config/profile.example.yaml config/profile.yaml

Edit config/profile.yaml. The LLM uses your proof points when generating solution hypotheses — the more concrete your profile, the sharper the output. See the example file for all available fields.

3. Set up Neon Postgres

Create a free Neon database, then run the schema migration:

npx drizzle-kit push

4. Configure environment variables

cp .env.example .env.local

Fill in your values. See .env.example for documentation on each variable.

5. Deploy to Vercel

Connect the repo to Vercel, add your environment variables in the dashboard, and deploy. Cron jobs run automatically on Vercel's Hobby tier at no cost.

Configuration

Signal Engine is driven entirely by config/profile.yaml. Key sections:

Section	Purpose
domain.strong	Terms matched against solicitations/reports for capability scoring
domain.exclusions	Keywords that immediately disqualify an item
capabilities.proof_points	Past contracts used by the LLM for concrete solution hypotheses
contract_preferences.naics_whitelist	NAICS codes to filter SAM.gov results (empty = no filter)
contract_preferences.size_range	Floor/ceiling contract values for hard-gate filtering
primary_agency_name	Substring matched against SAM.gov fullParentPathName to scope to your target agency

A well-filled profile produces useful triage cards. An empty profile produces generic noise.

Running locally

npm run dev

Trigger ingestion manually via the "Run Now" button in the dashboard, or call the cron endpoints directly:

curl -H "Authorization: Bearer $CRON_SECRET" http://localhost:3000/api/cron/scan-all
curl -H "Authorization: Bearer $CRON_SECRET" http://localhost:3000/api/cron/analyze-pending

Project structure

config/
  profile.example.yaml    Company profile template — copy to profile.yaml
  sources.yaml            Scraper configuration (OIG, GAO, SAM)
lib/
  fit-model/              Two-layer scoring: hard gates + LLM capability match
  analysis/               LLM prompt construction and triage card generation
  ingestion/              SAM.gov, OIG, GAO scrapers
  db/                     Drizzle schema and queries
app/
  (dashboard)/            Main dashboard UI
  api/cron/               Cron endpoint handlers
  api/feedback/           Thumbs up/down API

License

MIT