FomoFeed AI Orchestrator

What It Does

AI Orchestrator is the central brain that coordinates 6 different AI services in FomoFeed. When a post/moment is created, it runs all AIs and returns the best result.

---

Why 3 HF Spaces + 3 PHP Services?

HuggingFace Spaces (3)

Why HF? Only ML models, no database needed

1. CLIP Visual Analysis

   • Image quality analysis (0-10)
   • AI-generated suggested tags
   • Why HF? PyTorch model, GPU required

2. Timing Optimizer

   • Optimal hour prediction
   • CLIP boost + Video boost
   • Why HF? ML model, static computation

3. Feed Ranker

   • Engagement score prediction (0-100)
   • Why HF? ML model, feature engineering

PHP Backend (3)

Why PHP? Database access required, real-time data

4. User Profiler (api/ai/user_profiler.php)

   • Engagement history (engage_events table)
   • Follower count (users table)
   • Peak hours calculation
   • Why PHP? Database queries required!

5. Caption Generator (api/ai/suggest_caption.php)

   • Creates captions from CLIP tags
   • Why PHP? Already existing endpoint

6. Notification Generator (api/ai/notification_generator.php)

   • Smart notification text
   • Fetches username (users table)
   • Why PHP? Database required!

---

AI Flow

User: Presses "Analyze" button
    |
AI ORCHESTRATOR
    |
    v
STEP 1: User Profiler (PHP)
    engagement_hours=[19,20,21], follower_count=150
    |
    v
STEP 2: CLIP Visual (HF)
    quality_score=8.5, tags=["sunset", "nature"]
    |
    v
STEP 3: Timing Optimizer (HF)
    Input: engagement_hours + CLIP score 8.5
    Output: optimal_hour=19 (CLIP boost +30%)
    |
    v
STEP 4: Feed Ranker (HF)
    Input: CLIP=8.5 + Timing=19 + Follower=150
    Output: engagement_score=78
    |
    v
STEP 5: Caption Generator (PHP) - Optional
    Input: CLIP tags=["sunset", "nature"]
    Output: "Beautiful sunset colors..."
    |
    v
STEP 6: Notification Generator (PHP)
    Input: All AI results
    Output: "User is posting amazing content!"
    |
    v
RESULT: {
  "optimal_time": 19,
  "expected_reach": 780,
  "suggested_caption": "Beautiful sunset...",
  "confidence": 0.82
}

---

Data Flow

User Profiler
    -> Timing Optimizer (engagement_hours)
    -> Feed Ranker (follower_count, avg_engagement)

CLIP
    -> Timing Optimizer (quality_score -> CLIP boost)
    -> Feed Ranker (quality_score)
    -> Caption Generator (suggested_tags)

Timing Optimizer
    -> Feed Ranker (optimal_hour)

All
    -> Notification Generator (final recommendation)

Result: Each AI feeds into others -> 300% better results through coordination!

---

Deployment

1. Create HuggingFace Space

# Create new Space
https://huggingface.co/new-space

# Upload app.py, requirements.txt, Dockerfile
git clone https://huggingface.co/spaces/YOUR_USERNAME/ai-orchestrator
cd ai-orchestrator
cp ai_orchestrator_full.py app.py
git add .
git commit -m "Initial commit"
git push

2. Add Backend PHP Endpoints

# Copy to api/ai/ folder
cp user_profiler.php /path/to/api/ai/
cp notification_generator.php /path/to/api/ai/

3. Test

# Test in Postman
POST https://YOUR-ORCHESTRATOR-SPACE.hf.space/orchestrate
Body: {
  "user_id": 12,
  "content_type": "post",
  "media_path": "base64_image_here",
  "caption": "Test caption #ai",
  "tags": ["ai", "test"],
  "action": "analyze"
}

---

Test Checklist

• User Profiler -> GET /api/ai/user_profiler.php?user_id=12
• Notification Gen -> POST /api/ai/notification_generator.php
• CLIP -> POST https://bravo24-clip-ai-service.hf.space/analyse
• Timing -> POST https://bravo24-timing-optimizer.hf.space/predict
• Feed Ranker -> POST https://bravo24-feed-ranker.hf.space/predict
• Orchestrator -> POST /orchestrate (ALL AIs should work)

---

FAQ

Why some AIs in HF, others in PHP?

HF Spaces:

• Run ML models
• No database needed
• Static computations
• GPU required (for CLIP)

PHP Backend:

• Database access needed
• Real-time data fetching
• User-specific information
• Existing infrastructure

Why other AIs not in Orchestrator?

Quality Scorer: Auto-runs in create_post.php
Auto-CLIP: Auto-runs in create_post.php
Session Tracker: Frontend (Home Tab)
Creator Insights: Separate analytics dashboard

Orchestrator only coordinates AIs that run during "analyze" moment.

Is Orchestrator required?

No! Orchestrator is optional. Current system already works:

• analyze_content.php calls Timing + Feed Ranker directly
• Orchestrator is for advanced coordination in the future

---

Performance Improvement

---

Troubleshooting

"User Profiler error"

Check if user_profiler.php is installed

"CLIP timeout"

First request takes 30-60 seconds (cold start)

"All AI systems used: null, null"

Check backend PHP endpoints

---

v2.0.0 - Full AI Coordination System