backend/services/model_server.py

```python
import xgboost as xgb
import pandas as pd
import numpy as np
from fastapi import APIRouter
from ta.momentum import RSIIndicator
from ta.trend import MACD, EMAIndicator
from ta.volatility import BollingerBands, AverageTrueRange
from ta.volume import OnBalanceVolumeIndicator

router = APIRouter()
model = xgb.XGBClassifier()
model.load_model("models/model_xgb_v1.json")

def compute_rsi(close_prices, window=14):
    rsi_indicator = RSIIndicator(close=close_prices, window=window)
    return rsi_indicator.rsi()

def compute_technical_indicators(df):
    # Trend indicators
    ema8 = EMAIndicator(close=df['close'], window=8)
    ema21 = EMAIndicator(close=df['close'], window=21)
    df['EMA_8'] = ema8.ema_indicator()
    df['EMA_21'] = ema21.ema_indicator()
    df['EMA_cross'] = (df['EMA_8'] > df['EMA_21']).astype(int)
    
    # Momentum indicators
    df['RSI_14'] = compute_rsi(df['close'])
    macd = MACD(close=df['close'])
    df['MACD'] = macd.macd()
    df['MACD_signal'] = macd.macd_signal()
    df['MACD_hist'] = macd.macd_diff()
    
    # Volatility indicators
    atr = AverageTrueRange(high=df['high'], low=df['low'], close=df['close'], window=14)
    df['ATR_14'] = atr.average_true_range()
    
    bb = BollingerBands(close=df['close'])
    df['Bollinger_width'] = (bb.bollinger_hband() - bb.bollinger_lband()) / df['close']
    
    # Volume indicators
    df['Volume_change'] = df['volume'].pct_change()
    obv = OnBalanceVolumeIndicator(close=df['close'], volume=df['volume'])
    df['OBV'] = obv.on_balance_volume()
    
    # Price shape features
    df['candle_body'] = df['close'] - df['open']
    df['upper_shadow'] = df['high'] - df[['open', 'close']].max(axis=1)
    df['lower_shadow'] = df[['open', 'close']].min(axis=1) - df['low']
    
    # Lag returns
    for n in [1, 3, 5]:
        df[f'return_{n}'] = df['close'].pct_change(periods=n)
    
    # Time features
    df['minute_sin'] = np.sin(2 * np.pi * df.index.minute / 60)
    df['minute_cos'] = np.cos(2 * np.pi * df.index.minute / 60)
    
    return df

@router.post("/api/signal/{symbol}")
def get_signal(symbol: str, candles: list[dict]):
    df = pd.DataFrame(candles)
    df = compute_technical_indicators(df)
    latest = df.tail(1)
    
    features = ['EMA_8', 'EMA_21', 'EMA_cross', 'RSI_14', 'MACD', 'MACD_signal', 
                'MACD_hist', 'ATR_14', 'Bollinger_width', 'Volume_change', 'OBV',
                'candle_body', 'upper_shadow', 'lower_shadow', 'return_1', 
                'return_3', 'return_5', 'minute_sin', 'minute_cos']
    
    X = latest[features]
    prob_up = float(model.predict_proba(X)[0,1])
    prob_down = 1 - prob_up
    confidence = abs(prob_up - prob_down)

    advice = "long_bias" if prob_up > 0.6 and confidence > 0.15 else "short_bias" if prob_down > 0.6 else "neutral"
    
    return {
        "symbol": symbol,
        "prob_up": round(prob_up, 3),
        "prob_down": round(prob_down, 3),
        "confidence": round(confidence, 3),
        "advice": advice
    }
```