hire-signal

Automated job board monitor that polls Greenhouse, Ashby, and Lever career pages, scores listings with Claude AI, and sends interactive Discord notifications for high-fit engineering roles. Runs autonomously on GitHub Actions with Turso as the cloud database.

Architecture

GitHub Actions (cron schedule + manual dispatch)

  ┌──────────┐   ┌──────────┐   ┌──────────┐
  │ monitor  │ → │  score   │ → │  notify  │
  │ (ingest) │   │ (Claude) │   │(Discord) │
  └────┬─────┘   └────┬─────┘   └────┬─────┘
       │               │              │
       ▼               ▼              ▼
  ┌─────────────────────────┐  ┌──────────────────────┐
  │   Turso (hosted SQLite) │  │ Discord Bot API      │
  │   (local SQLite in dev) │  │ #job-alerts (bot)    │
  └─────────────────────────┘  │ #pipeline-errors (wh)│
                               └──────────────────────┘
                                        ▲
                               ┌────────┴─────────┐
                               │ Cloudflare Worker │
                               │ (button handler)  │
                               └──────────────────┘

Pipeline Steps

Step	Description
Monitor	Poll Greenhouse/Ashby/Lever APIs, keyword filter, staleness filter (skip jobs >14 days old), prune stale data, store new jobs in SQLite/Turso
Score	Score unscored jobs via Claude API on role fit, location, stack, and comp
Notify	Send Discord embeds with age badges and interactive "Seen" buttons for jobs scoring 7+
Alert	Post to `#pipeline-errors` via webhook if any step fails (used by CI)

Schedule

Runs on GitHub Actions cron:

Mon-Wed: 8 AM and 1 PM CT
Thu-Fri: 8 AM CT
Sunday: 6 PM CT
Plus manual dispatch via workflow_dispatch

Discord Integration

Job notifications are sent via a Discord Bot (not webhooks) to support interactive "Seen" buttons. Each message includes one button per company — clicking it marks all jobs for that company as seen (checkmark prefix on embed fields, button disappears).

A Cloudflare Worker handles button interactions: verifies Discord's ed25519 signature, updates Turso (seen_at timestamp), and edits the original message.

Error alerts use a separate webhook to #pipeline-errors — no bot or interactivity needed.

Monitored Companies

446 companies across three ATS platforms:

Greenhouse (265): Stripe, Coinbase, Airbnb, Figma, Discord, Robinhood, Anthropic, DoorDash, Pinterest, Databricks, and 255 more
Ashby (148): Notion, Ramp, Linear, OpenAI, Cursor, Docker, Snowflake, Plaid, 1Password, Vanta, and 138 more
Lever (33): Metabase, Crypto.com, Palantir, Match Group, Quizlet, WHOOP, Outreach, and 26 more

See config/companies.json for the full list. Companies are tagged with optional sector metadata (e.g., ai, saas, infrastructure, consumer-finance).

Adding a New Company

Find the board slug and ATS platform (see CLAUDE.md for detailed steps):
- Greenhouse: https://boards-api.greenhouse.io/v1/boards/{slug}/jobs
- Ashby: https://api.ashbyhq.com/posting-api/job-board/{slug}
- Lever: https://jobs.lever.co/v0/postings/{slug}?mode=json
Add to config/companies.json:
```
{ "name": "Rocket Money", "id": "truebill", "seed": false, "source": "greenhouse", "sector": "consumer-finance" }
```
- seed: false — scores all current jobs <14 days old on first run (recommended for new companies)
- seed: true — seeds existing jobs silently on first run (use for companies you already watch)
- source — which ATS platform the company uses ("greenhouse", "ashby", or "lever")
- sector — optional category tag (e.g., "ai", "saas", "infrastructure", "consumer-finance")
Either way, subsequent runs automatically only process new postings — no manual changes needed.
Commit and push — the next GitHub Actions run will pick it up automatically. Or trigger a manual run via workflow_dispatch.

How It Works

Seed vs. evaluate: Companies marked seed: true get their existing catalog silently ingested on first run (is_seed = 1) — those jobs are never scored or notified. Use this for companies you already watch. Companies with seed: false get all matching fresh jobs (<14 days) scored on first run. Either way, after the first run the system is self-managing: jobExists dedup ensures only genuinely new postings flow through the pipeline.

Keyword filtering: Job titles are matched against include keywords (e.g. software engineer, frontend, react, typescript) and exclude keywords (e.g. manager, data scientist, intern). Exclude takes priority.

Staleness filter: Jobs posted more than 14 days ago are skipped (not scored) since response rates drop significantly after the first two weeks. They are still tracked in the database for deduplication.

DB pruning: Runs automatically at the start of each monitor cycle. Job descriptions are nulled after 30 days (reclaims storage) and jobs are soft-archived after 90 days (excluded from scoring and notifications, but still used for dedup). Thresholds are configurable via STALENESS_THRESHOLD_DAYS, DESCRIPTION_PRUNE_DAYS, and ARCHIVE_DAYS in config.ts.

Getting Started

Prerequisites

Node.js 24+ (managed via Volta)
Yarn 1.x

1. Fork, clone, and install

git clone https://github.com/YOUR_USERNAME/hire-signal.git
cd hire-signal
yarn install

2. Customize your scoring profile

Edit profile/scoring-brief.md with your own details:

Role target (title, seniority, location preference)
Target comp range
Tech stack (strongest to familiar)
Experience and domains
Preferences and anti-patterns

The AI scorer uses this file as context when evaluating every job. Be specific — the more detail you provide, the more accurate scoring will be.

3. Customize your companies and keywords

Edit config/companies.json to set your target companies (see Adding a New Company).

Edit src/shared/config.ts to adjust keywords:

INCLUDE_KEYWORDS — title keywords to match (e.g., software engineer, frontend)
EXCLUDE_KEYWORDS — title keywords to filter out (e.g., manager, intern)

4. Set up external services (all free tier)

Anthropic (required for scoring):

Sign up at console.anthropic.com
Create an API key → ANTHROPIC_API_KEY

Discord Bot (required for notifications):

Create an app at discord.com/developers/applications
Create a Bot, copy the token → DISCORD_BOT_TOKEN
Note the Application ID → DISCORD_APP_ID and Public Key → DISCORD_PUBLIC_KEY
Invite the bot to your server: https://discord.com/api/oauth2/authorize?client_id={APP_ID}&permissions=83968&scope=bot
Copy the channel ID for notifications → DISCORD_CHANNEL_ID

Discord Webhook (required for error alerts):

In your Discord server, create a webhook in a #pipeline-errors channel → DISCORD_ERROR_WEBHOOK_URL

Turso (required for cloud/CI deployment):

Sign up at turso.tech
Create a database: turso db create job-monitor
Get the URL: turso db show job-monitor --url → TURSO_DATABASE_URL
Create a token: turso db tokens create job-monitor → TURSO_AUTH_TOKEN

5. Configure environment

cp .env.example .env
# Fill in your credentials

For local development, set DB_BACKEND=local (uses SQLite, no Turso needed).

6. Run locally

# Run the full pipeline
yarn monitor    # Ingest jobs from all company boards
yarn score      # Score unscored jobs with Claude AI
yarn notify     # Send Discord notifications for jobs scoring 7+

# Or run all three in sequence
yarn run-all

7. Deploy to GitHub Actions

Add these secrets to your repo (Settings → Secrets → Actions):

Secret	Description
`ANTHROPIC_API_KEY`	Claude API key
`DISCORD_BOT_TOKEN`	Discord bot token
`DISCORD_CHANNEL_ID`	Discord channel for notifications
`DISCORD_WEBHOOK_URL`	Discord webhook (legacy, kept for compatibility)
`DISCORD_ERROR_WEBHOOK_URL`	Discord webhook for error alerts
`TURSO_DATABASE_URL`	Turso database URL
`TURSO_AUTH_TOKEN`	Turso auth token
`SCORING_BRIEF_B64`	Base64-encoded scoring brief (`cat profile/scoring-brief.md \| base64`)

Push to main — the pipeline runs automatically on the cron schedule or via manual dispatch.

8. Deploy Cloudflare Worker (optional — enables "Seen" buttons)

cd worker
yarn install
npx wrangler login
npx wrangler secret put DISCORD_PUBLIC_KEY
npx wrangler secret put DISCORD_BOT_TOKEN
npx wrangler secret put TURSO_DATABASE_URL
npx wrangler secret put TURSO_AUTH_TOKEN
npx wrangler deploy

Copy the deployed URL and set it as the Interactions Endpoint URL in Discord Developer Portal → General Information.

Without the worker, notifications still work — you just won't have interactive "Seen" buttons.

Discovering New Companies

A discovery script scans 15K+ ATS slugs (from data/ats-slugs/) and identifies companies with matching engineering roles:

yarn discover --us-only             # scan all platforms, US/remote only
yarn discover --platform ashby      # scan one platform
yarn discover --limit 500           # cap slugs per platform

Results are saved to data/discovery-results.json. Review the output and add selected companies to config/companies.json. The slug lists are sourced from job-board-aggregator.

Development

# Type check
npx tsc --noEmit

# Lint
yarn lint

# Format
yarn format

Tech Stack

Language: TypeScript (ESM)
Runtime: Node.js via tsx (no build step)
Database: SQLite via better-sqlite3 (local) / Turso via @libsql/client (cloud)
AI: Claude API — Sonnet 4 via @anthropic-ai/sdk
Notifications: Discord Bot API (interactive messages with buttons)
Error Alerts: Discord webhooks
Interaction Handler: Cloudflare Worker (ed25519 verification, Turso HTTP API)
CI/CD: GitHub Actions (cron schedule + error alerting)

Estimated Cost

Service	Monthly
GitHub Actions	$0 (public repo)
Turso	$0 (free tier)
Cloudflare Worker	$0 (free tier, ~100 req/day)
Discord Bot	$0
Claude API (Sonnet 4)	~$10-30
Total	~$10-30/month

Name		Name	Last commit message	Last commit date
Latest commit History 68 Commits
.claude		.claude
.github/workflows		.github/workflows
.vscode		.vscode
config		config
scripts		scripts
src		src
worker		worker
.env.example		.env.example
.gitignore		.gitignore
.prettierrc		.prettierrc
CLAUDE.md		CLAUDE.md
README.md		README.md
eslint.config.js		eslint.config.js
package.json		package.json
tsconfig.json		tsconfig.json
yarn.lock		yarn.lock

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

hire-signal

Architecture

Pipeline Steps

Schedule

Discord Integration

Monitored Companies

Adding a New Company

How It Works

Getting Started

Prerequisites

1. Fork, clone, and install

2. Customize your scoring profile

3. Customize your companies and keywords

4. Set up external services (all free tier)

5. Configure environment

6. Run locally

7. Deploy to GitHub Actions

8. Deploy Cloudflare Worker (optional — enables "Seen" buttons)

Discovering New Companies

Development

Tech Stack

Estimated Cost

About

Uh oh!

Releases

Packages

Uh oh!

Contributors

Uh oh!

Languages

Folders and files

Latest commit

History

Repository files navigation

hire-signal

Architecture

Pipeline Steps

Schedule

Discord Integration

Monitored Companies

Adding a New Company

How It Works

Getting Started

Prerequisites

1. Fork, clone, and install

2. Customize your scoring profile

3. Customize your companies and keywords

4. Set up external services (all free tier)

5. Configure environment

6. Run locally

7. Deploy to GitHub Actions

8. Deploy Cloudflare Worker (optional — enables "Seen" buttons)

Discovering New Companies

Development

Tech Stack

Estimated Cost

About

Topics

Resources

Uh oh!

Stars

Watchers

Forks

Releases

Packages 0

Uh oh!

Contributors

Uh oh!

Languages

Packages