Café com Dopamina

Overview

"What if you could distill the entire week's tech trends into a 5-minute read — every single day, automatically?"

Café com Dopamina is an automated content platform that aggregates data from multiple sources (GitHub trending, LinkedIn engagement, Twitter/X, newsletters), feeds it through an AI-powered pipeline, and publishes daily episode-style blog posts in Brazilian Portuguese.

The platform targets Brazilian developers, tech leads, and cloud engineers who want to stay on top of global tech trends without doom-scrolling multiple feeds.

🎯 Key Objectives

📊 Aggregate trending tech data from 5+ sources daily 🤖 Use LLMs to generate high-quality, opinionated content in Portuguese 🔄 Fully automated pipeline — data collection → content generation → PR → publish 🌐 SEO-first Next.js site with server-rendered pages and structured data ☕ Authentic voice — like talking to a friend over coffee

🏗️ Technical Architecture

┌─────────────────────────────────────────────────┐
│              DATA COLLECTION                     │
│                                                  │
│  GitHub API  ─→  trending repos & devs          │
│  LinkedIn    ─→  high-engagement posts          │
│  Twitter/X   ─→  trending discussions           │
│  RSS/HN      ─→  newsletter highlights          │
└──────────────────────┬──────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────┐
│           CONTENT GENERATION                     │
│                                                  │
│  build_episode_digest.py  → structured JSON     │
│  generate_episode.py      → LLM (GitHub Models) │
│  episode_prompt.txt       → Portuguese voice     │
└──────────────────────┬──────────────────────────┘
                       │
                       ▼
┌─────────────────────────────────────────────────┐
│              PUBLISHING                          │
│                                                  │
│  GitHub Action opens PR in blog repo            │
│  Human review → merge → Vercel auto-deploy      │
│  Cross-post to portfolio blog                   │
└─────────────────────────────────────────────────┘

💻 Data Pipeline

# Daily at 14:00 UTC — GitHub Action
# 1. Collect last 24h of data from all sources
python scripts/build_episode_digest.py --days 1

# 2. Generate episode via GitHub Models API (GPT-4o)
python scripts/generate_episode.py

# 3. Push to blog repo + open PR for review
# 4. Cross-post adapted version to portfolio blog

🔑 Technical Highlights

• GitHub Models API — Uses GITHUB_TOKEN for LLM calls, zero extra secrets
• Cross-repo automation — Single workflow pushes content to 2 separate repos
• Anti-detection scraping — LinkedIn data collected via Playwright with full stealth stack
• SEO-first — Server-rendered, JSON-LD structured data, dynamic OG images via Satori
• Markdown-driven — Episodes are plain .md files with YAML frontmatter, parsed at build time
• Human-in-the-loop — AI generates, but every episode is reviewed before publish

📈 Impact

• 70+ data sources monitored daily (GitHub, LinkedIn, RSS, Reddit, HN)
• Episodes generated in ~30 seconds end-to-end
• Zero manual data collection — fully automated pipeline
• Content published in Brazilian Portuguese for an underserved audience