📄 tiered-pipeline-ui-notes.md 4,698 bytes Apr 21, 2026 📋 Raw

Tiered Pipeline UI Notes — For Daedalus 🎨

From: Socrates 🧠
To: Daedalus 🎨
Date: 2026-04-21
Status: ✅ BACKEND WIRED AND READY

Quick Status Check

Before building the UI, verify the Gaming PC is up:

curl http://localhost:8000/admin/content/health

Expected response:

{
  "gaming_pc": {"available": true, "reason": "Ollama responding"},
  "models": {
    "strategy": "llama3.1:8b",
    "structure": "qwen2.5-coder:7b",
    "draft": "phi4:14b",
    "seo": "llama3.1:8b"
  },
  "pipeline_ready": true
}

The Pipeline (Live on Gaming PC)

Stage Model Role Latency
1. Strategy Llama 3.1 8B Brief + angle ~10s
2. Structure Qwen 2.5 7B Validate angle before burning GPU time ~15s
3. Draft Phi-4 14B Full 800-1200 word draft ~60-90s
4. SEO Llama 3.1 8B Excerpt, tags, meta ~10s
5. Compliance Python Strip banned words, flag dates/names ~1s

Total: ~90-120 seconds (varies with Phi-4 generation speed)

The bottleneck is Stage 3 (Phi-4). Everything else is fast.

Magic Wand Button — Loading State

Since Stage 3 takes 60-90s, the UI needs clear progress indication:

Required UX:
- Clear spinner/progress bar
- Current stage name (e.g., "Drafting with Phi-4...")
- Estimated time remaining (seconds)
- Disable button while running
- Allow cancellation
- Show warnings if compliance flags issues

Suggested pattern:

[🪄 Generate Content] → click

↓

[⠋ Generating... Stage 3/5: Drafting with Phi-4 (~45s remaining)]
[━━━━━━━━━━━░░░░░] 60%
[Cancel]

API Endpoints (Ready)

1. Health Check

GET /admin/content/health

Check if Gaming PC Ollama is available before allowing generation.

2. Start Generation

POST /admin/content/generate
Content-Type: application/json

{
  "topic": "How I migrated from Google Calendar to CalDAV",
  "content_type": "how_i_solved",
  "outline": ["optional", "bullet", "points"],
  "context": "Additional notes"
}

Response:

{
  "job_id": "gen_20260421_171500_a1b2c3",
  "status": "queued",
  "estimated_seconds": 90
}

3. Poll for Status

GET /admin/content/jobs/{job_id}

Poll every 2-3 seconds. Returns:

{
  "job_id": "gen_20260421_171500_a1b2c3",
  "status": "draft",
  "stage_number": 3,
  "total_stages": 5,
  "progress_percent": 60,
  "current_stage_name": "Drafting with Phi-4...",
  "estimated_seconds_remaining": 45,
  "started_at": "2026-04-21T17:15:00",
  "completed_at": null,
  "error": null,
  "result": null
}

When complete, result contains the full output:

{
  "job_id": "gen_20260421_171500_a1b2c3",
  "status": "completed",
  "progress_percent": 100,
  "result": {
    "title": "The Night I Broke DNS",
    "content": "# The Night I Broke DNS\n\nIt was 7 PM...",
    "excerpt": "A DNS migration gone wrong and how I fixed it",
    "tags": ["dns", "tailscale", "homelab"],
    "meta_description": "How I migrated DNS and broke everything",
    "word_count": 1150,
    "reading_time_minutes": 6,
    "compliance": {
      "replacements_made": 2,
      "warnings": ["Found: 'leverage' → replaced with 'use'"],
      "is_compliant": true,
      "banned_words_found": ["leverage"],
      "dates_flagged": [],
      "names_flagged": []
    }
  }
}

4. Cancel Job

POST /admin/content/jobs/{job_id}/cancel

Content Types

  • roundup — Collection/review post
  • how_i_solved — Problem + solution narrative
  • build_log — Step-by-step project documentation
  • essay — Opinion/analysis piece
  • tutorial — Instructional content

Default is how_i_solved.

Error States

Error HTTP Status UI Display
Gaming PC offline 503 "Gaming PC offline. Check Tailscale or retry later."
Model timeout 504 "Generation timed out. Retry with shorter outline."
Compliance flags 200 Show warnings but allow use (yellow banner)
Job not found 404 "Job expired. Start over."

Files for Reference

  • Backend: hoffdesk-api/content/pipeline.py (the orchestrator)
  • Admin templates: shared/project-docs/blog/templates/admin/
  • Admin CSS: shared/project-docs/blog/admin.css

Testing

Want to see it work? SSH to the Beelink and run:

cd /home/hoffmann_admin/.openclaw/workspace-socrates/hoffdesk-api
python test_pipeline_live.py "Your topic here"

This actually hits the Gaming PC and takes ~90 seconds.

Backend is ready. Over to you for the Magic Wand UI.

— Socrates 🧠

← Back