Daily AI Learning Journal

A zero-touch pipeline that publishes a daily entry of me learning Claude AI in public, posted to six platforms every morning at 10:30 AM IST — without me touching the laptop.

The goal: prove that a non-technical operator can use Claude to build, ship, and run real software end-to-end. The journal itself is the proof.

What runs every day

06:00 IST  cron in GitHub Actions
             ↓
  ┌──────────────────────────────────────────────────────────────────┐
  │  pipeline/research.py    gather day's Claude news from Reddit /  │
  │                          HackerNews / web / GitHub trending      │
  │                                  ↓                                │
  │  pipeline/writer.py      Gemini Flash (primary) / Claude (fallback)│
  │                          → article, twitter thread, IG caption,  │
  │                          LinkedIn post, Threads post              │
  │                                  ↓                                │
  │  pipeline/assets/*       Pillow renders 5-card carousel + cover   │
  │  pipeline/assets/video   FFmpeg + edge-tts → 60s vertical video   │
  │                                  ↓                                │
  │  git commit + push       to main; site rebuilds on GitHub Pages  │
  │                                  ↓                                │
  │  pipeline/publishers/    parallel posts: dev.to, Hashnode,        │
  │                          Twitter, Threads, Instagram, LinkedIn   │
  └──────────────────────────────────────────────────────────────────┘
             ↓
  publish_report.json with all 6 URLs committed back to repo

Runs entirely on the GitHub Actions free tier (2000 min / month). One daily run uses ~3 min.

Stack

Layer	Tool	Why
Orchestrator	GitHub Actions cron	Free, reliable, no server to babysit
Content gen	Gemini 1.5 Flash (primary), Claude Sonnet 4.6 (fallback)	Gemini is free-tier; Claude better at voice consistency
Research	Serper (web), Reddit JSON, HN API, GitHub trending	All free / generous free tiers
Image assets	Pillow + DejaVu fonts	No paid design tool; deterministic output
Video	FFmpeg + edge-tts	Free Microsoft voice; vertical 1080×1920 for IG/Reels
Publishing	dev.to API, Hashnode GraphQL, Twitter v2, Threads API, instagrapi, LinkedIn REST	Each platform's free tier is enough for one post/day
Static site	GitHub Pages	Auto-deploys from `docs/`

Repo layout

pipeline/
  research.py              gather sources
  writer.py                LLM call + voice/style guardrails
  assets/card_generator.py 5-card carousel + cover image
  assets/video_generator.py 60s vertical narrated video
  publishers/              one module per platform
  main.py                  orchestrator
.github/workflows/daily.yml cron + xvfb headless display
docs/index.html            auto-regenerated catalog page
templates/ig_profile.png   profile circle for cards
requirements.txt
.env.example

Voice + safety guardrails

The writer's system prompt enforces:

No buzzwords ("leverage", "synergy", "paradigm" are blocked)
No AI-sounding phrases ("Certainly!", "As an AI", "I apologize")
Short sentences; Indian English is fine
Never mention employer names, internal projects, or confidential work — explicit in the prompt
Always include one copy-paste example a beginner can try the same day

Running locally

# Clone the repo, then:
cd claude-daily-journal
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env       # fill the API keys for whichever platforms you want
python -m pipeline.main

Only ANTHROPIC_API_KEY (or GEMINI_API_KEY) is required. Every publisher is independent — skip the ones whose env vars are missing.

Why this exists

I'm a Customer Success operator, not a developer. Most people who built things like this had backgrounds in engineering. I wanted the journal to be a daily public proof that you don't.

Co-built with Claude as pair-programmer over many iterations.

MIT licensed.