AI Pulse

Automated AI newsletter system that curates and delivers twice-weekly digests for builders, solo founders, product managers, and vibe coders.

How It Works

15+ Sources (RSS / Scrape / Reddit / Google News / Hacker News)
        ↓
   Aggregator  →  Fetch & parse articles from dynamic + static sources
        ↓
     Curator   →  LLM Filter → Dedupe → LLM Categorize → Rank → Generate treatments
        ↓
    Delivery   →  Render HTML/text → Send emails via Resend
        ↓
     Archive   →  Store digests as HTML files

Smart Curation (not just keywords)

The system uses LLM-based intelligence at two critical stages:

1. Content filtering — The LLM classifies each article as include/exclude based on whether it's actionable technical content. Opinion pieces, regulatory news, lawsuits, and corporate financials are filtered out automatically.

2. Category assignment — Instead of keyword matching, the LLM uses its knowledge of the AI ecosystem to categorize articles. "Gemma 4" gets categorized under Google without needing the word "google" in the article. "Kiro" maps to AWS. "Llama 4" maps to Meta.

Both stages fall back to keyword-based heuristics if the LLM is unavailable.

Article Treatment

Every article that passes curation gets a 2-layer treatment:

• Why It Matters — implications for builders and product development
• What You Can Build — 3-4 concrete, actionable use cases

Coverage Categories

Category	Examples
Anthropic/Claude	claude, sonnet, opus, haiku, claude code
OpenAI	gpt, chatgpt, dall-e, sora, o1, o3, o4, codex, whisper
Google	gemini, gemma, deepmind, vertex, veo, imagen
AWS	bedrock, sagemaker, titan, amazon q
Meta AI	llama, codellama
Builder Tools & OSS	cursor, copilot, windsurf, kiro, antigravity, deepseek, mistral, grok, langchain, ollama, huggingface, and 30+ more

Setup

# Install dependencies
npm install

# Copy environment variables
cp .env.example .env

Edit .env with your keys:

RESEND_API_KEY=re_your_resend_api_key_here
LLM_API_KEY=sk-your-llm-api-key-here
ARCHIVE_BASE_PATH=./archive

The default LLM provider is Groq (llama-3.3-70b-versatile). You can switch to Anthropic or OpenAI by modifying the config.

Usage

Start the scheduler

Runs the pipeline automatically on Monday and Friday at 6:00 AM CST:

npm start

CLI commands

# Manually trigger the newsletter pipeline
npx tsx src/cli.ts run-pipeline

# Manage sources
npx tsx src/cli.ts list-sources
npx tsx src/cli.ts add-source <name> <url> <type> <categories>
npx tsx src/cli.ts remove-source <id>

# Manage subscribers
npx tsx src/cli.ts list-subscribers
npx tsx src/cli.ts add-subscriber <email>
npx tsx src/cli.ts remove-subscriber <id>

# View archived digests
npx tsx src/cli.ts list-archive

Source types: rss, atom, scrape, tool-radar

Adding a Subscriber

To manually add an email address to the newsletter:

npx tsx src/cli.ts add-subscriber user@example.com

This creates a subscriber with pending status and sends a confirmation email via Resend. The subscriber must click the confirmation link to start receiving newsletters. Once confirmed, their status changes to active.

To view or remove existing subscribers:

npx tsx src/cli.ts list-subscribers
npx tsx src/cli.ts remove-subscriber <id>

A default subscriber is seeded automatically on first run (configured in src/config.ts).

GitHub Actions

The included workflow (.github/workflows/newsletter.yml) runs the pipeline on a cron schedule (Monday & Friday at 12:00 UTC). It can also be triggered manually via workflow_dispatch.

Required repository secrets: RESEND_API_KEY, LLM_API_KEY.

Testing

# Run all tests
npm test

# Run unit tests only
npm run test:unit

# Watch mode
npm run test:watch

Tech Stack

• TypeScript + Node.js
• SQLite (better-sqlite3) — article storage, subscriber management, delivery logs
• Resend — email delivery
• Groq / Anthropic / OpenAI — LLM-based content curation and categorization
• rss-parser + cheerio — feed parsing and web scraping
• node-cron — scheduling
• vitest — testing

License

Private.