# Architecture Research: Content Automation Pipeline **Researched:** 2026-04-05 ## System Overview ``` ┌─────────────────────────────────────────────────────────────────┐ │ n8n (Orchestration) │ │ │ │ ┌──────────┐ ┌──────────┐ ┌──────────┐ ┌──────────┐ │ │ │ AI Idea │ │ News/RSS │ │ Client │ │ Market │ │ │ │ Generator│ │ Scraper │ │ Pipeline │ │ Intel │ │ │ └────┬─────┘ └────┬─────┘ └────┬─────┘ └────┬─────┘ │ │ │ │ │ │ │ │ └─────────────┴────────────┴─────────────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Content │ │ │ │ Pool │◄── NocoDB │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Script │ │ │ │ Generator │──── Claude API │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Human │ │ │ │ Review │◄── NocoDB UI │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Video │ │ │ │ Generate │──── HeyGen API │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Post- │ │ │ │Production │──── Creatomate API │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Client │ │ │ │ Approval │◄── NocoDB UI │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Publish │ │ │ │ Instagram │──── Meta Graph API │ │ └─────┬─────┘ │ │ │ │ │ ┌─────▼─────┐ │ │ │ Analytics │ │ │ │ Tracker │──── Instagram Insights API │ │ └───────────┘ │ └─────────────────────────────────────────────────────────────────┘ ``` ## Component Boundaries ### 1. n8n Workflows (Orchestration Layer) **Separate workflows per concern** (not one monolithic workflow): | Workflow | Trigger | Purpose | |----------|---------|---------| | `idea-generator-ai` | Cron (daily 8am) | AI generates content ideas from client profile | | `idea-generator-news` | Cron (every 6h) | Pull RSS, filter, score, insert ideas | | `idea-generator-market` | Cron (weekly Mon) | Pull Destatis/Bundesbank data, create angles | | `idea-generator-client` | NocoDB webhook | Client adds listing -> auto-create content idea | | `script-generator` | Manual trigger / n8n trigger | Take approved idea -> Claude API -> draft script | | `video-generator` | n8n trigger (after script approval) | Script -> HeyGen API -> wait for render -> download | | `post-production` | n8n trigger (after video ready) | Raw video -> Creatomate template -> branded video | | `review-notifier` | NocoDB webhook (status change) | Notify team/client when content needs review | | `publisher` | Cron (scheduled times) | Check for approved+scheduled content -> publish | | `token-refresh` | Cron (every 50 days) | Refresh Instagram long-lived token | | `analytics-collector` | Cron (daily) | Pull Instagram Insights for posted content | **Why separate:** Each workflow can fail independently, be debugged independently, and be version-controlled independently. Monolithic workflows become unmaintainable. ### 2. NocoDB Schema (Data Layer + Client UI) **Tables:** ``` clients ├── id (auto) ├── name (text) ├── brand_kit_path (text) — path to brand assets on VPS ├── instagram_account_id (text) ├── instagram_token (text, hidden) ├── avatar_configs (JSON) — HeyGen avatar IDs ├── content_strategy (JSON) — categories, split, tone └── active (boolean) content_ideas ├── id (auto) ├── client_id (link -> clients) ├── source (single select: ai_generator | news | client | market) ├── category (single select: property | market_insight | company | educational) ├── title (text) — brief idea title ├── description (long text) — full idea description ├── data_points (JSON) — any stats/numbers to include ├── source_url (URL) — original article/data source ├── relevance_score (number 0-10) — AI-assigned ├── status (single select: new | selected | scripted | rejected | duplicate) ├── created_at (date) └── content_hash (text) — for dedup checking scripts ├── id (auto) ├── idea_id (link -> content_ideas) ├── client_id (link -> clients) ├── avatar (single select: owner | employee) ├── format (single select: talking_head | avatar_visuals | avatar_broll) ├── instagram_format (single select: reel | feed | story) ├── script_text (long text) — the full German script ├── caption (long text) — Instagram caption ├── hashtags (text) ├── hook (text) — first 3 seconds ├── cta (text) — call to action ├── visual_notes (long text) — what visuals to show alongside ├── duration_seconds (number) ├── internal_status (single select: draft | internal_review | approved | needs_changes) ├── client_status (single select: pending | approved | needs_changes) ├── internal_reviewer (text) ├── client_notes (long text) └── approved_at (date) generated_videos ├── id (auto) ├── script_id (link -> scripts) ├── client_id (link -> clients) ├── heygen_job_id (text) — for polling/webhook ├── raw_video_path (text) — unbranded video file ├── branded_video_path (text) — post-production video file ├── thumbnail_path (text) ├── creatomate_render_id (text) ├── status (single select: generating | raw_ready | branding | branded_ready | failed) ├── generation_cost (number) — track API credits used ├── error_message (text) └── created_at (date) published_posts ├── id (auto) ├── video_id (link -> generated_videos) ├── client_id (link -> clients) ├── instagram_post_id (text) ├── instagram_format (single select: reel | feed | story) ├── published_at (datetime) ├── scheduled_for (datetime) ├── caption (long text) ├── hashtags (text) ├── status (single select: scheduled | published | failed) └── publish_error (text) analytics ├── id (auto) ├── post_id (link -> published_posts) ├── client_id (link -> clients) ├── collected_at (date) ├── likes (number) ├── comments (number) ├── shares (number) ├── saves (number) ├── reach (number) ├── impressions (number) ├── engagement_rate (number) — calculated └── category (single select) — from content_ideas for performance analysis content_calendar ├── id (auto) ├── client_id (link -> clients) ├── slot_date (date) ├── slot_time (time) ├── day_of_week (single select) ├── category (single select) — recurring theme ├── assigned_content_id (link -> scripts or generated_videos) └── status (single select: open | assigned | published) ``` ### 3. File Storage (VPS) ``` /var/www/neg-marketing/ ├── brand-kits/ │ └── neg/ │ ├── logo.svg │ ├── logo-white.svg │ ├── fonts/ │ ├── colors.json │ └── templates/ — Creatomate template configs ├── avatars/ │ └── neg/ │ ├── owner-source.mp4 — original recording │ └── employee-source.mp4 ├── generated/ │ └── neg/ │ └── 2026-04/ │ ├── raw/ — HeyGen output │ ├── branded/ — Creatomate output (final) │ └── thumbnails/ ├── assets/ │ └── neg/ │ ├── properties/ — property photos from client │ ├── b-roll/ — location footage │ └── music/ — royalty-free tracks └── temp/ — processing workspace ``` ### 4. Data Flow ``` [Content Sources] ──► [NocoDB: content_ideas] ──► [Human selects] │ ▼ [Claude API] ◄── [n8n: script-generator] ──► [NocoDB: scripts] │ [Human edits/approves] │ [Client approves] │ ▼ [HeyGen API] ◄── [n8n: video-generator] ──► [VPS: raw video] │ ▼ [Creatomate API] ◄── [n8n: post-production] ──► [VPS: branded video] │ ▼ [n8n: publisher] ──► [Instagram via Meta Graph API] │ ▼ [n8n: analytics] ──► [NocoDB: analytics] ``` ### 5. Review/Approval Flow via NocoDB NocoDB status fields drive the workflow. n8n watches for status changes via webhooks: ``` Script created (draft) → n8n notifies internal reviewer → Reviewer opens NocoDB, reads script, sets status: - "approved" → n8n triggers client notification - "needs_changes" → n8n notifies script author → Client opens NocoDB, reads script, sets status: - "approved" → n8n triggers video generation - "needs_changes" → n8n notifies team with client notes ``` **No custom review UI needed.** NocoDB's form view + gallery view provides a clean approval interface. Client sees only their content via filtered views. ## Suggested Build Order ``` Phase 1: Foundation ├── NocoDB setup (tables, views, API) ├── n8n setup (base configuration) ├── File storage structure on VPS ├── Brand kit upload and template design └── Meta Developer App + Instagram Business setup Phase 2: Content Generation Pipeline ├── AI Idea Generator workflow (n8n + Claude API) ├── News/RSS ingestion workflow ├── Client input pipeline (NocoDB webhook) ├── Content dedup logic └── Script generation workflow (n8n + Claude API) Phase 3: Video Production Pipeline ├── HeyGen avatar creation from recordings ├── Video generation workflow (n8n + HeyGen API) ├── Creatomate branded template design ├── Post-production workflow (n8n + Creatomate) └── Review/approval workflow Phase 4: Publishing + Analytics ├── Instagram publishing workflow ├── Scheduling system (content calendar + cron) ├── Analytics collection workflow ├── Token refresh workflow └── Performance feedback into content strategy ``` **Dependencies:** - Phase 2 depends on Phase 1 (NocoDB + n8n must exist) - Phase 3 depends on Phase 2 (need scripts to generate videos) + Phase 1 (brand templates) - Phase 4 depends on Phase 3 (need videos to publish) - Content generators (Phase 2) can partially overlap with Phase 3 --- *Researched: 2026-04-05*