Cosmic Digest

A personal news aggregator that sends you a daily email with tech news you actually care about. Uses GPT to summarize the day's headlines and throws in a coding challenge to keep you sharp.

I built this because most tech newsletters are either too broad or too niche. This one learns your interests and filters accordingly.

What it does

• Pulls from RSS feeds (Hacker News, TechCrunch, etc.)
• Scores articles based on your topics/keywords/regions
• GPT-5.1 writes a summary with the important bits
• Adds a daily coding problem (algorithm, system design, or practical stuff)
• Emails you a clean digest every morning
• Runs for free on GitHub Actions

Example digest

Daily Digest — Tuesday, December 31, 2024 at 8:00 AM EST

## AI Summary
- Next.js 15.0: React Server Components are now the default, new App Router caching, 40% faster builds with Turbopack. Pages Router needs manual opt-in now. [1]
- Claude 3.7 Sonnet: 200K context (doubled), 88% on SWE-bench. Available via API. [3]

## Worldwide but relevant to you
- TypeScript 5.6 Beta Announcement — Microsoft (2024-12-30)
- The State of WebAssembly 2025 — Hacker News (2024-12-29)

## Challenge of the Day
Implement an LRU cache with O(1) get and put operations.
Difficulty: Medium | Skills: Hash Tables, Doubly Linked List

Tech

• .NET 10
• Resend for emails (free tier: 100/day)
• OpenAI GPT-5.1 for summaries (~$0.02 per digest)
• Markdig for markdown → HTML
• GitHub Actions for scheduling (free)

Setup

You need .NET 10 SDK, a Resend API key, and an OpenAI API key.

git clone <your-repo>
cd Cosmic-Digest

# Install .NET 10 if needed (WSL/Ubuntu)
wget https://dot.net/v1/dotnet-install.sh
chmod +x dotnet-install.sh
./dotnet-install.sh --channel 10.0

# Config
cp .env.example .env
# edit .env with your keys

dotnet restore
dotnet run

You'll get an email right away.

Configuration

Edit .env:

# Required
RESEND_API_KEY=re_xxxxx
OPENAI_API_KEY=sk-proj-xxxxx
MAIL_TO=you@example.com
TIMEZONE=America/New_York

# Personalization
PREF_TOPICS=ai,frontend,backend,open source,devops
PREF_KEYWORDS=React,TypeScript,Python,Docker,Kubernetes,Next.js
PREF_REGIONS=US,Silicon Valley,Canada

# Sources
RSS_FEEDS=https://hnrss.org/frontpage,https://techcrunch.com/feed/,...

# Optional
ENABLE_AI_SUMMARY=true
PRICE_WATCH=M2 MacBook Air|https://example.com|USD

For full-stack devs: Topics: ai, frontend, backend, web development, devops, open source Keywords: React, Next.js, TypeScript, Node.js, Docker, PostgreSQL, CI/CD

For ML engineers: Topics: ai, machine learning, deep learning, NLP Keywords: LLM, GPT, PyTorch, transformers, fine-tuning, RAG

GitHub Actions (automated daily emails)

The workflow is already set up in .github/workflows/daily-digest.yml.

1. Add your secrets to GitHub repo settings (Settings → Secrets → Actions):

   • RESEND_API_KEY
   • OPENAI_API_KEY
   • MAIL_TO
   • TIMEZONE
   • PREF_TOPICS
   • PREF_KEYWORDS
   • PREF_REGIONS
   • RSS_FEEDS

2. Push to GitHub

3. It runs daily at 8 AM UTC. You can change the cron schedule or trigger it manually from the Actions tab.

See GITHUB_ACTIONS_SETUP.md for details.

How it works

Relevance scoring: Each article gets points for matching your preferences:

• Keywords: +1.0 each
• Topics: +0.6 each
• Regions: +0.4 each
• Recency boost: 0-0.5 (decays over 3 days)

Only articles scoring >0.75 make it to your digest.

AI summaries: GPT-5.1 reads the top 15 articles and writes 6-8 bullets. I tuned the prompt to force specific version numbers, metrics, and citations. The citation numbers are clickable links to the source articles.

Daily challenges: Rotates between algorithms, data structures, system design, and practical problems. Same challenge all day (deterministic based on date), new one tomorrow. Should take 15-30 minutes.

Price tracking: Naive HTML scraping for now. Checks if price is above/below recent average. You can plug in real APIs later.

Project structure

Cosmic-Digest/
├── Program.cs              # Main pipeline
├── NewsAi.cs              # GPT summarization
├── ChallengeGenerator.cs  # Daily challenges
├── DigestComposer.cs      # Email builder
├── RssIngestor.cs         # RSS fetcher
├── Relevance.cs           # Scoring algorithm
├── PriceTracker.cs        # Price monitoring
├── StateStore.cs          # JSON persistence
├── Models.cs              # Data models
├── .env                   # Your config (gitignored)
└── data/state.json        # Article cache (tracked in git for Actions)

Cost

• GitHub Actions: Free (2,000 min/month)
• Resend: Free (100 emails/day)
• GPT-5.1: ~$0.60/month for daily digests

Total: less than a dollar a month.

Why I built this

Most news feeds are noise. I wanted something that:

• Learns what I care about
• Doesn't spam me with 50 headlines
• Actually helps me stay current in tech
• Keeps my coding skills fresh

The relevance scoring works surprisingly well. The AI summaries save time. The daily challenges are a nice habit.

Tradeoffs

JSON file vs Database

I'm using a JSON file (data/state.json) for persistence instead of a database. This works because:

• The dataset is small (few hundred articles max)
• No concurrent writes (single daily job)
• GitHub Actions can commit state changes back to the repo
• Zero hosting costs

If you need to scale to thousands of users or run more frequently, switch to PostgreSQL or SQLite. The StateStore.cs abstraction makes this swap easy.

GitHub Actions vs Cloud hosting

Running on GitHub Actions instead of a dedicated server or cloud function. The tradeoff:

• ✅ Completely free (2,000 min/month)
• ✅ No infrastructure to manage
• ✅ State persists via git commits
• ❌ Cold starts on every run
• ❌ Must commit state to repo (pollutes git history)
• ❌ 2,000 minute monthly limit

For more frequent digests (hourly, real-time), use Vercel Cron or AWS Lambda with a real database. For personal daily use, GitHub Actions is perfect.

GPT API vs Local LLM

I'm using OpenAI's GPT-5.1 API instead of running a local model. Here's why:

• GPT-5.1 quality is significantly better for technical summaries
• Costs ~$0.60/month (negligible for personal use)
• No GPU or hosting infrastructure needed
• API is reliable and fast

If privacy is critical or you're processing thousands of digests, run Llama 3 or Mistral locally. You'll need a GPU and the quality won't be as good, but it's free and private.

Known issues

• Price scraping is brittle (HTML parsing always is)
• No web UI yet (everything is env vars)
• Citations sometimes break in certain email clients
• State file grows unbounded (should add cleanup)

Roadmap

• Web dashboard for config
• Replace price scraping with actual APIs
• Article sentiment analysis
• Database instead of JSON
• Browser extension for reading list

Pull requests welcome.

Security

Don't commit your .env file. Use GitHub Secrets for automation. If you accidentally expose API keys (like I did), revoke them immediately.

License

MIT