AI_CONFIGURATION.md

# 🤖 AI Configuration Guide

## 🔧 **Fixed Issues:**

### ✅ **Real AI Integration (No More Mock Responses)**

The chat API now uses **real AI** instead of template responses:

1. **Trained Model Priority**: Uses your trained model if available
2. **HuggingFace API Fallback**: Uses HuggingFace Inference API
3. **Smart Mock Fallback**: Only uses mock responses as last resort

### ✅ **AI Response Flow:**

```
User Question → Trained Model → HuggingFace API → Smart Mock
```

## 🚀 **How It Works:**

### **1. Trained Model (Best)**
- If `./models/textilindo-trained` exists
- Uses your custom-trained model
- Provides most accurate responses

### **2. HuggingFace API (Good)**
- Requires `HUGGINGFACE_API_KEY` environment variable
- Uses powerful pre-trained models
- Good for general questions

### **3. Smart Mock (Fallback)**
- Only when AI models fail
- Improved keyword matching
- Better response relevance

## 🔑 **Setup HuggingFace API Key:**

### **Option 1: Environment Variable**
```bash
export HUGGINGFACE_API_KEY=your_api_key_here
```

### **Option 2: HuggingFace Space Settings**
1. Go to your space settings
2. Add environment variable: `HUGGINGFACE_API_KEY`
3. Set value to your HuggingFace API key

### **Get API Key:**
1. Visit: https://huggingface.co/settings/tokens
2. Create new token with "Read" permissions
3. Copy the token

## 🧪 **Test AI Integration:**

### **Test Questions:**
```bash
# Test 1: Operating hours
curl -X POST "https://harismlnaslm-Textilindo-AI.hf.space/chat" \
  -H "Content-Type: application/json" \
  -d '{"message": "Jam berapa Textilindo buka?"}'

# Test 2: Location
curl -X POST "https://harismlnaslm-Textilindo-AI.hf.space/chat" \
  -H "Content-Type: application/json" \
  -d '{"message": "dimana lokasi textilindo?"}'

# Test 3: Shipping
curl -X POST "https://harismlnaslm-Textilindo-AI.hf.space/chat" \
  -H "Content-Type: application/json" \
  -d '{"message": "apa ada gratis ongkir?"}'
```

## 📊 **Expected Results:**

### **Before (Mock Responses):**
```json
{
  "message": "Jam berapa Textilindo buka?",
  "response": "Textilindo berkantor pusat di Jl. Raya Prancis...",
  "status": "success"
}
```

### **After (Real AI):**
```json
{
  "message": "Jam berapa Textilindo buka?",
  "response": "Jam operasional Senin-Jumat 08:00-17:00, Sabtu 08:00-12:00.",
  "status": "success"
}
```

## 🎯 **Training Your Model:**

### **1. Start Training:**
```bash
curl -X POST "https://harismlnaslm-Textilindo-AI.hf.space/api/train/start" \
  -H "Content-Type: application/json" \
  -d '{
    "model_name": "distilgpt2",
    "dataset_path": "data/lora_dataset_20250910_145055.jsonl",
    "max_samples": 20,
    "epochs": 1
  }'
```

### **2. Monitor Progress:**
```bash
curl "https://harismlnaslm-Textilindo-AI.hf.space/api/train/status"
```

### **3. Test Trained Model:**
```bash
curl -X POST "https://harismlnaslm-Textilindo-AI.hf.space/api/train/test"
```

## ✅ **Verification:**

### **Check AI Status:**
```bash
curl "https://harismlnaslm-Textilindo-AI.hf.space/health"
```

### **Test Chat:**
```bash
curl -X POST "https://harismlnaslm-Textilindo-AI.hf.space/chat" \
  -H "Content-Type: application/json" \
  -d '{"message": "Jam berapa Textilindo buka?"}'
```

## 🎉 **Success Indicators:**

- ✅ **Relevant Responses**: AI answers match the questions
- ✅ **No More Mock Fallbacks**: Real AI responses
- ✅ **Trained Model Priority**: Uses your custom model
- ✅ **HuggingFace API**: Works with API key
- ✅ **Smart Fallbacks**: Better error handling

---

**Your AI is now using real intelligence instead of templates! 🚀**