AI-Automated Tech Shorts Platform

1) Solution Overview

This platform automates a faceless tech-news YouTube Shorts channel. It fetches tech sources, generates scripts, creates voice + captions, composes short-form video, validates policy compliance, and uploads on schedule.

2) Architecture

RSS Feeds / APIs
      |
      v
[Ingest + Dedupe] ---> SQLite(job queue, runs, uploads, metrics)
      |
      v
[LLM Script Gen]
   |   \
   |    \__ Primary: Gemini API
   |       Fallback: Local Ollama (qwen2.5:1.5b)
   v
[TTS Audio] --> [FFmpeg Render + Captions + Safe Zones] --> [Compliance Gate]
                                                           |
                                                           v
                                                [YouTube Upload + Schedule]
                                                           |
                                                           v
                                                 [Dashboard + Metrics Pull]

3) End-to-End Workflow

1. Ingest: Pull RSS feeds and rank by relevance/recency.
2. Generate: Produce title, 60-second script, description, hashtags.
3. TTS: Convert script to audio.
4. Render: Build 1080x1920 MP4 with captions and safe-margin overlays.
5. Validate: Block risky/policy-sensitive content and high-similarity scripts.
6. Upload: Push validated jobs to YouTube with retries.
7. Observe: Monitor statuses/errors in dashboard and logs.

4) Infrastructure & Domains

• schulabs.com: primary website endpoint on Nginx + Let's Encrypt.
• dashboard.schulabs.com: reverse-proxy to internal dashboard service on 127.0.0.1:8088.
• imagemaster.schulabs.com: reserved public app slot.
• systemd timers run ingest/generate/render/validate/upload/metrics/cleanup.

# Timer check
systemctl list-timers --all | grep pipeline

# Upload logs
journalctl -u pipeline-upload.service -n 120 --no-pager

5) Security Model

• SSH key-based access for deploy user.
• Firewall (UFW) and fail2ban enabled.
• TLS certificates via Certbot.
• Dashboard auth via WEB_USERNAME / WEB_PASSWORD.
• Secrets stored server-side in /etc/tech-shorts/pipeline.env and /opt/tech-shorts/secrets/.

6) Operations Runbook

# Manual single-item run
sudo -u techops bash -lc 'cd /opt/tech-shorts && source .venv/bin/activate && pipeline generate && pipeline render && pipeline validate && pipeline upload'

# Dashboard service
systemctl status pipeline-web.service --no-pager -l

# Restart full automation stack
systemctl restart pipeline-generate.timer pipeline-render.timer pipeline-validate.timer pipeline-upload.timer

7) Cost Strategy

Design target is near-zero incremental spend during early validation:

• Gemini free-tier as primary script model.
• Automatic local fallback model when cloud limit is hit.
• Local FFmpeg rendering.
• YouTube API within normal quota envelope for 2/day upload schedule.

8) Is this anonymous and fully hands-off?

Short answer: faceless yes, fully anonymous no.

• Faceless: yes, channel can avoid showing your identity publicly.
• Platform anonymity: no, YouTube/Google account ownership and API credentials are traceable to account owner.
• Hands-off: mostly, but not zero-maintenance; periodic checks are still required for policy, quality, and quota limits.

9) Black Screen Issue: Root Cause & Fix

Earlier videos appeared as black screen + text when no background media was available. The renderer now includes:

• Source-image extraction from article OG/Twitter meta images.
• Pexels fallback (if API key present).
• Safe-zone overlay positioning to avoid Shorts UI overlap.
• Improved text styling and readable caption zone.

Recommendation: keep a branded default visual in /opt/tech-shorts/assets/default_bg.jpg to guarantee non-empty backgrounds.

10) Troubleshooting

11) Next Milestones

1. Finalize branded visual pack (intro, lower-third, default backgrounds).
2. Tune compliance threshold and script diversity prompts to reduce blocks.
3. Add weekly auto-report page (views, uploads, failures, top topics).
4. Deploy ImageMaster app to imagemaster.schulabs.com.